9YardsTerminal
Wed, Jul 29 ETWire
@chamathhttps://t.co/baZ2XV4BHI· 27d@JasonMy team at The Syndicate is meeting with family offices to review some very interesting deals we're working on.
If you would like to take a meeting, please email fo@launch.co· 27d@JasonThe appearance of impropriety that the 47th administration is signaling is the perfect way to accelerate the democratic socialist movement, the America First/America Only movement & the creation of @TuckerCarlson's 3rd party.
We need to get money out of politics by paying the President a top salary, say the median/mean salary of the top publicly traded companies ($25M a year/$100m a term).
We then need to do something similar (at a smaller scale) for VPs and members of Congress.
Finally, we have to limit all donations and cap SuperPACs' spending.
Our system is completely bought and sold at this point.
Please explain to me why you think I'm wrong.· 27d@Jason⚔️ @jalenbrunson1 ⚔️ https://t.co/3kxGMQgJ9k· 27d@eladgilSource of question -
https://t.co/9BRcsFknOL· 27d@xaiGrok Build is now installed in Railway sandboxes· 27d@HarryStebbingsThe single best follow in tech on X right now is @nikesharora.
One of the best operators of our time spewing wisdom on a daily basis.· 27d@KTmBoyle“This man seemed enthusiastic and ambitious, but I must confess to holding some serious doubts about the dramatic scope of his plans and his ability to pull them off… well, he did the thing.”
This is the story arc. Aspire to this. 🇺🇸🚀💪· 27d@PalmerLuckeyBe careful with your fireworks this weekend!· 27d@saranormoussuper impressive example of combining machines-asking-humans in order to efficiently direct use of compute against a goal· 27d@chamathyup.· 27d@packyMThis is real btw https://t.co/WOyE5WLijL· 27d@zebulgarSince everyone is sharing interesting World Cup stats
My interesting one, is that so far, every winning team
Has been the one that has scored the most goals in their game· 27d@packyMThe next 250 years of the American project should be unimaginably abundant, if we can keep it.
Read Scott Nolan's vision for America's Next 250 in today's Not Boring. https://t.co/FzdY5ySngJ· 27d@HarryStebbingsDario just declared war on open-source.
Anthropic's message is clear: open source could destroy the entire AI business model, and Chinese open-source models are the cause.
I sat down with @jasonlk & @rodriscoll to discuss it, along with the biggest news in tech this week:
- Anthropic & Dario Declare War on Open-Source
- Coinbase Slashes AI Spend 50%: Is the Token Bubble Bursting?
- Kalshi's $40BN Valuation & Impending IPO
- Bending Spoons: The Smartest IPO of 2026 & the $100BN SaaS Roll-Up Play
My notes below:
1. Anthropic Is Laying the Foundation for a Deal With the U.S. Government on Chinese Models
Anthropic accused Chinese open-source AI competitors of “brazen theft” through model distillation that violates its terms of service. Rory noted the hypocrisy, given that U.S. frontier models originally scraped external IP. Still, Anthropic appears to be laying the groundwork for a regulatory trade: complying with domestic access restrictions in exchange for a federal ban on distilled Chinese models.
2. Jason Lemkin’s Response to Brian Armstrong’s AI Tweet
Jason dismissed Coinbase CEO Brian Armstrong’s 50% LLM cost-cutting post as “performative social media,” arguing that savings matter little if core revenue is flat or shrinking. He believes AI must actively drive top-line growth. Rory defended the move as “cost management 101,” saying cash-conscious enterprise executives will quickly emulate it to curb runaway frontier model fees.
3. CEOs Are Struggling to See the ROI From AI
Massive enterprise spending on AI tokens is failing to deliver the expected revenue or productivity gains, leaving CFOs searching for measurable operational lift. Jason noted that adding millions in AI spend can still produce the same growth rates as prior quarters. Rory argued that AI spending must clearly accelerate software delivery or create definitive bottom-line savings as boards push back on reckless “token maxing.”
4. We Are All So Aligned in Wanting AI to Win
Jason warned that the U.S. economy is structurally addicted to AI, with 40% of the S&P 500 tied to the boom, making society eager to prop it up to protect 401(k) portfolios. Rory countered that protectionism artificially inflates intelligence costs for the broader economy. He compared it to banning IBM PC clones in the 1980s just to protect IBM’s stock price, calling the blocking of low-cost open-source alternatives “fricking dumb.”
5. Bending Spoons and the New Playbook for B2B Revenue Arbitrage
Bending Spoons’ $20 billion public valuation marks a shift toward tech roll-ups that drive profitability through price hikes and cost-cutting rather than organic user growth. Jason predicts this playbook will expand into mature B2B SaaS. By acquiring sticky but underperforming platforms like Marketo, Asana, or PagerDuty and injecting hungry talent, operators can rapidly improve retention and capture massive revenue arbitrage.
(links below)· 27d@garrytanPlease oppose ACA-7
If you live in California take a moment to click the link here and tell your representatives
We want an end to racism of all kinds in our schools· 27d@chamathPart of what 8090 will be able to show, over time, is that companies powered by us generate return on capital employed (ROCE) that exceed their peers.
We help companies do this by encoding their "edge" into software that can recursively improve vs leaving it as tribal knowledge or tacit conventions that sit in people's heads but nowhere else. Because when those who know what to do leave a company, it creates a hole and those that remain often find themselves treading water or moving backwards as a result.
From a recent BCG report:
"As the cost of capital has returned to long-term averages (8% to 11%), roughly half of large US companies cannot deliver returns that exceed that.
Persistently low returns on capital (as measured by return on capital employed, ROCE) are surprisingly common across public companies, affecting about one in seven companies globally.
These underperforming businesses represent a widespread challenge confronting many portfolios. Even healthy businesses grapple with divisions that underperform, dragging down overall profitability and complicating strategic decision-making."· 27d@altcapTake me home! 🇺🇸🇺🇸🇺🇸 https://t.co/aWWYyI7Isr· 27d@packyMIf I were FIFA, I’d think very carefully about what happens to haughty Europeans who try to impose rules we don’t like on us without our consent on our home soil around July 4th. https://t.co/4KQYnQ5htZ· 27d@pmarcaInteresting.· 27d@pmarcaInteresting.· 27d@pmarcaMany such cases.· 27d@levieIf you’ve ever wondered why we will need 100X more AI inference in the future, and what it’s going to be driven by, this is another good example.
Devin pushes forward an idea of agentic mapreduce, which means we’ll now have swarms of agents that are processing large amounts of data (code) to handle tasks that humans never could have done before.
“Devin maps relevant signals across the repo, fans out focused agents over bounded shards, reduces their findings into one report, then verifies serious vulnerabilities in isolated sandboxes before marking them confirmed.”
In this case it’s code security, but there are tons of other use-cases in code and knowledge work. We see this at Box with customers that want to process and understand millions of documents for risk, insights, relationships, and more. This will play out in pharma, banking, and many other industries across all forms of unstructured data.
As an aside, these types of capabilities are generally only possible when you can deploy a variety of models (both the frontier and lower cost) because of the sheer amount of tokens that go into these use-cases. This is going to be a major value proposition for the applied AI layer.· 28d@zebulgarAMERICAAAAA FUCK YEAHHHH
2-0
🇺🇸🇺🇸🇺🇸 https://t.co/oyjppKLLf7· 28d@garrytanhttps://t.co/gTrjV7mcxf· 28d@martin_casado“In the near term, some routine tasks like coding and debugging will fall back to Opus 4.8” 😳🫠· 28d@saranormousnuclear powered GPUs!!· 28d@perplexity_aiClaude Fable 5 is once again available in Computer as an orchestrator model. https://t.co/U65gnvja30· 28d@altcapThe American Dream has never belonged to one party. It belongs to every child. Grateful to be in Newark w @SenBooker signing up kids for @TrumpAccounts. In NJ over 1 M kids have at least $250 waiting in their free, lifetime investment account. 🇺🇸🚀 https://t.co/QIPzh7f0aw· 28d@martin_casadoIt's baaaaaaaaaccckkkk https://t.co/jP4sk1A8zA· 28d@garrytanMost underreported story in all of tech· 28d@Lightspeedvp.@bsomaia on AI valuations:
“Are we in a super cycle? Yes.”
“Are we in a very rich valuation environment? Yes.”
“Will there be investments that despite this valuation environment will be extremely rewarding? Yes.”
"Will there be investments that turn out to lose money? Yes."
"All of those things can be true."
From Startup Policy Forum with @ShwetaRKohli.· 28d@cohereHappy Canada Day!! 🇨🇦
We're so proud to be founded and growing in Canada. Let's keep building💡· 28d@bgurleyWeChat has been a little quiet on AI front & perhaps even falling into being considered a laggard. But this week, Pony Ma presented a new vision on how to turn "mini-apps" (still no real equivalent in US) into AI handshakes. Manny partners announced. Could be a powerful move. Read more below.· 28d@LightspeedvpThe team at @unconvAI has released their first model, Un-0!
Un-0 is an image-generation model powered by a simulated system of coupled oscillators.
Instead of the massive arithmetic operations that power most AI, the model uses simulated physics. Thousands of virtual oscillators influence each other, sync up, and settle into patterns. Feed it an input, let the oscillators interact, and the resulting pattern is the output.
On ImageNet 64×64, Un-0 achieves an FID score of 6.74, matching the quality of leading conventional image generation methods when they were first published.
The company has stated that, to their knowledge, this is the most capable image generator to date that uses a simulation of a physical dynamical system.
Unconventional AI believes energy efficiency demands a fundamentally different computer, one where physics does the computing. We believe this is a meaningful milestone in their goals of achieving ~1,000x energy-efficiency gains in modern AI.
Congratulations to @NaveengRao, @MeeLan_Lee, @sachour__, @MCarbin, and the entire Unconventional AI team. We are incredibly excited for what’s ahead.
CC: @guruchahal, @ravi_lsvp, and @buckymoore· 28d@xaiIntroducing Voice Agent Builder: a no-code platform to create human-like voice agents with Grok Voice.
Available today at $0.05 / min.
https://t.co/kUkF7zqvfR https://t.co/OCIq1oDYar· 28d@LightspeedvpA big congratulations to the @Arena team on hitting $100M ARR!
This milestone announcement comes just eight months after the launch of their evaluation product.
Arena hosts millions of users who can work in real time with AI agents, testing and comparing different AI models. One primary feature is their AI model performance leaderboard, which has been generated from over 10 million user evaluations.
Incredible milestone @ml_angelopoulos, @infwinston, @istoica05, and team. We’re proud to see Arena continuing to develop and advance the frontier of AI through community-driven evaluation.
Learn more in the post below.· 28d@eladgil🔥 @harvey on fire· 28d@accelOur 2026 Founders Factory report is now live!
Europe and Israel's startup ecosystem has entered a new phase. Employees from 322 VC-backed unicorns have gone on to found more than 2,300 startups across the region, raising over $42B and creating a self-sustaining ecosystem where talent, experience and capital compound with each generation of founders.
At the same time, AI is accelerating this momentum, creating a new generation of companies that are scaling from zero to unicorn status in record time.
In this year's report, produced in partnership with @dealroomco and @RevelioLabs, we explore:
➡️ How more than 2,300 startups have emerged from Europe and Israel's unicorn founder factories
➡️ How AI is creating Europe’s fastest-ever unicorns, with 1 in 5 new unicorns since 2023 reaching a $1B valuation in under 2 years
➡️ Which European startup ecosystems are creating the most founder factories, with London producing more than any other city while Sweden’s founder factories generate nearly 3 times more startups per unicorn than the European average
➡️ How the archetype of Europe’s newest unicorn founders is changing, with today’s founders more likely to come from Big Tech and academia
Read the full report here ⬇️ https://t.co/sqAjYX47CD· 28d@cohereThe future is yours to build.
Cohere x @AstonMartinF1 https://t.co/K2P5iJJ6ue· 28d@KTmBoyleEndless summer. https://t.co/vXtZeWSGww· 28d@altcapThank u to Sec Lutnick, @NotTomBrown & @WhiteHouse for moving so quickly to get America’s frontier models back online. Finding the safety balance was never going to be easy - but America is winning & this rapid response & new framework will accelerate us yet again. 🇺🇸🚀· 28d@levieThings seem to be ending up in a better spot with Fable, and presumably GPT-5.6 next. What we have now is the initial precedent for what frontier model releases (or at least those that have significant coding and cyber capabilities) could look like going forward. This would presumably apply to bio and other categories of risk that have been identified by AI safety groups.
From the Anthropic post:
“3. A shared industry framework. Although we have reached a constructive resolution, these events have made clear that the industry needs a consistent way to assess and fix potential “jailbreaks” of AI models (techniques that bypass a model’s safeguards).2 A shared standard for judging the severity of a given jailbreak would help AI developers triage new findings as they arise, launch highly-capable models with greater safety, and communicate the level of risk consistently to government and industry partners. Together with Amazon, Microsoft, Google, and other Glasswing partners, we’ve started to develop such a framework, and we outline it below.
4. Deeper government collaboration. We’re also strengthening our level of collaboration with the US government on new pre-release testing, information sharing, and research collaboration. We describe this deeper collaboration in the final section.”
It’s been a messy process to get here, but at least there’s some semblance of a framework that could be practical. The only note of caution here would be that there’s a lot of subjectivity that goes into various risks and their actual levels of exploitability in practice. We’re likely going to be living with a framework that requires heavy judgment and back and forth between labs and the government for major releases.
The best we can hope for is that this is a relatively efficient process, and hopefully has ways of being sped up for incremental version updates in models. It would be a bad outcome if every release after this level of threshold of capability required the same review process, and we don’t get the same rate of breakthroughs we’ve been seeing.· 29d@AnthropicAIClaude Fable 5 will be available again globally tomorrow.
After a series of productive conversations with the US government, we're redeploying the model with a new set of classifiers to target and block more cybersecurity tasks. In the near term, some routine tasks like coding and debugging will fall back to Opus 4.8. We’ll continue to refine these classifiers over the coming weeks to reduce false positives and better distinguish genuine misuse from legitimate requests.
We’ve also begun drafting a consensus framework—with Amazon, Microsoft, Google, and other Glasswing partners—for assessing the severity of AI jailbreaks and how AI developers should respond to them. We invite other industry partners and model providers to join us in this effort.
Finally, we’re scaling up our collaboration with the US government on model testing and safeguards. This will include pre-release access to models and safeguards for evaluation, information sharing on jailbreaks and misuse, and dedicated resources for joint research.
Thank you to our users for your patience, and to our partners across the government, industry, and the research community who worked alongside us to make Fable 5 available again.
Read our full blog: https://t.co/VHyum831ri· 29d@bgurleyThis is absolutely correct. The only way to fight "open" is with "open." Glad to hear a voice in Washington that understands the real stakes.· 29d@AnthropicAIWe’ve received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5.
We'll begin restoring access tomorrow, and will share an update soon.
We’re grateful to our users for their patience, and to everyone who worked with us on redeploying the models.· 29d@HarryStebbings"After food, energy sovereignty is one of the most important strategic priorities for any nation.
History shows that wars are often fought over access to essential resources, and today, energy is at the center of that competition.
The more countries can produce reliable energy domestically, the more resilient, secure, and economically independent they'll become."
What is more important energy sovereignty or model sovereignty @jameswise @vkhosla @alanchanguk @wolfejosh @traestephens· 29d@martin_casadoHey book lovers.
I'm re-reading Hesse's Steppenwolf, and it's an entirely different book than when I read it in my 20s. The perspective of age changes everything.
What other books are like this? Worth a re-read a few decades later to get a wholey new perspective on?· 29d@cohereWe build differently than other tech firms.
No superintelligence. No competition to spend the most money. We're creating empowering, efficient AI to enhance human potential, not replace it.
With high standards for engineering & innovation, we're proud to support @AstonMartinF1. https://t.co/q5ZoRNV9Qe· 29d@levieWe've been running Anthropic's Claude Sonnet 5 through the Box AI Complex Work Eval, our agentic benchmark that puts models through real enterprise document work end-to-end.
Sonnet 5 holds frontier-class quality on complex multi-step work and pulls ahead of Sonnet 4.6 in several core enterprise domains like Energy (+4.7pp), Retail (+4.4pp), and Professional Services (+2.6pp), and other spaces where unstructured data is heavily complex.
Here are a few examples of wins compared to Sonnet 4.6 to get a sense of some of the more advanced reasoning capabilities in Sonnet 5:
* Financing due diligence: It computed the company's liquidity and leverage ratios from the raw balance sheet, and caught that the source report's own stated debt-to-equity figure understated the leverage, flagging all three loan covenants as violated, not just the ones the document admitted.
* Overhaul cost analysis: It scoped "total cost" to the company's own KPI definitions, correctly separating out Lost Production Cost because the guidance said to track it separately rather than naively summing every number on the sheet. It also caught and handled a broken reference cell in the spreadsheet.
* SKU revenue analysis: On segmented sales data, it computed each product's contribution against the correct subcategory denominator, sidestepping the easy mistake of dividing by the category total, and flagged why no Pet-category SKU cracked the top 9.
Sonnet 5 will be available in the Box AI Studio shortly for customers to build custom agents with.· 29d@a16zVacation looks very different depending on where you work.
Over the course of the year, Germans hold the OOO crown. But not consistently! Swedes and Italians trade places in July and August, and in August, the French take over. Brazil breaks from the European pattern on summer break length. India books vacation days at the last minute.
And birthdays are popular days off, but only the big ones. Turning 31? Not worth celebrating, apparently.
We teamed up with @deel to look at how startup and tech workers around the world actually take vacation, not just how many days they're offered, but when they take them, how long, and whether they take them at all.
North America is stingier than Europe. Most "vacations" are a single day off. Long weekends spike on Mondays and Fridays in summer.
And no matter how hard the Europeans work to make their summer holidays count, Christmas break wins everywhere. The British are proportionately at their desks in peak summer; the Dutch over Christmas.
Except Armenia, for some reason.
Full piece: https://t.co/UpFUh4KTEF· 29d@perplexity_aiClaude Sonnet 5 is now available in Perplexity for Pro and Max subscribers.
You can also select it as an orchestrator model in Computer. https://t.co/UktzCrUZU6· 29d@saranormousMany parts of venture don’t make any sense to founders, and hanging out with the great but always surprising @stanine, being forced to explain how VC works, reminded me of that· 29d@zebulgarEv and Jack thanked me like 5 times for not overly bashing Benchmark in this
I'm not sure if that means this podcast is good or banal, but hey we had some fun· 29d@OpenAIWe’re introducing GeneBench-Pro, a research-level benchmark for a harder kind of AI progress: how well agents can navigate messy biological data, choose the right analysis path, and make judgment calls that real computational research depends on.
https://t.co/AsilnnSxnE· 29d@GoogleDeepMindWe’re shipping 2 major releases:
🔘 Nano Banana 2 Lite: our fastest and cheapest Gemini Image model
🔘 Gemini Omni Flash: now available via the Gemini API and in @GoogleAIStudio to help developers generate and edit high-quality videos.· 29d@bgurleySuper cool momentum here. Groundbreaking public/private partnerships. And the most effective program for inequality — direct wealth transfer. No unnecessary rake. Great to see new incremental use cases. Love this - business supporting local community.· 29d@generalcatalystWe backed @1001ai_ at seed and are continuing to back the team in their $30M Series A.
1001 builds sovereign AI for the critical infrastructure that moves the physical world: aviation, ports, and energy. It's built and governed locally, so operators stay in control.
Congratulations to @b_abughazaleh, Mohammed Fituri, and the entire team.· 29d@KTmBoyle“The problem isn’t freedom. It’s that we’ve lost the primary institution where virtues are formed— the household in which children form one another.” https://t.co/98j4nWHxUO· 30d@accelThe most important professional conversations and the best ideas rarely happen in front of a screen. They happen in the moments between meetings, over the phone and across tables. Yet many of the tools we rely on to capture them were built for a desk-bound world that doesn't reflect how most people actually work. As a result, follow-ups are forgotten, information stays siloed, and great ideas can slip away.
@Heypocket solves this. Built by co-founders @AkshayNarisetti and @gabrieldymowski, Pocket is a dedicated AI device that helps people stay present. Elegant enough that you don’t have to remember to bring it with you and capable enough that it captures everything you need it to. Joining the Pocket journey as we participate in the Seed round, we’re looking forward to supporting the team as they reimagine the way people reflect, recall and work.
Welcome to the Accel family, Pocket! Read more from Cecilia Wang and Matt Robinson ⬇️ https://t.co/SwJqxGBIPn· 30d@xaiState of the art voice APIs from SpaceXAI, now in the Vercel AI Gateway· 30d@a16zAirbnb, 2008. https://t.co/uBQZL8BEUF· 30d@a16zThe a16z Editorial team’s summer reading list: https://t.co/RxvtUDs2xr https://t.co/lWyb54IyJf· 30d@PalmerLuckeyAt least 1,200 people died of heatstroke in France last Wednesday, mostly elderly people indoors.
Even crazier, about 70,000 Europeans die from the same every year, akin to a bi-weekly 9/11. Eurobros, please vote for leaders who will let you open your windows and install A/C! https://t.co/NxK4irX7GG· 31d@eladgilStatus
https://t.co/e9pZtfyPGr· 32d@fintechjunkieThe Bear was……perfect.· 32d@AnthropicAISince June 12, we’ve been working closely with the US government to restore access to Claude Mythos 5 and Fable 5. Today, the government notified us that Mythos 5, our strongest cybersecurity model, can be redeployed to a set of US organizations that operate and defend critical infrastructure.
We’re restoring access for these organizations quickly, and we’re continuing to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again.· 33d@OpenAIIntroducing a limited preview of GPT-5.6 Sol, our next generation frontier model, as well as GPT-5.6 Terra, a balanced model for efficient, everyday work, and GPT-5.6 Luna, a fast and affordable model for high-volume work.
https://t.co/OoM83SyISN· 33d@PalmerLuckeyJust two days until this incredible vehicle goes up for auction! https://t.co/tzCUYBaHyC· 34d@typesfastTIL the Caesar salad was invented in Tijuana. Mexican food really is the best.· 34d@typesfastIs Rippling going to eat every SaaS category?· 34d@sequoiaToday's AI models train once. We don't work that way. We learn continuously, forget what doesn't matter, and retain what does.
That gap is what @dan_biderman and @realJessyLin are closing at @EngramLab. AI that never stops learning, with memory that lives inside the model instead of bolted on as an afterthought.
In our latest Training Data episode we get into why memory is the next frontier: why the brain forgets on purpose, why RAG is a band-aid, and what becomes possible when a model is always training.
00:00 Introduction
00:59 Always Training Explained
01:51 Beyond Context Windows
03:29 Ngram Product Overview
04:34 Adapters And Training Signals
05:32 Internalize Vs Externalize
06:49 Compute And Token Savings
08:19 Teams First Then Individuals
08:51 Memorization Vs Understanding
12:47 Dreams And Offline Digestion
14:08 Training Beats Curation
15:19 Why Everyone Needs A Model
21:44 Bitter Lesson And Architecture
24:44 RAG Killer And KV Cache
31:38 Future Of Memory And Models· 34d@OpenAIWork at OpenAI is being transformed by agents, in every department.
Across our entire company, people are using Codex to do work that is more complex, longer-running, and increasingly cross-functional.
Our internal usage offers an early look at how agentic tools may reshape work as they become more capable and broadly available.· 34d@typesfastWow JSX just added Santa Monica to Oakland service· 34d@GoogleDeepMindGemini 3.5 Flash now supports native computer use.
This built-in tool lets developers build custom agents that can see and take action across browser, mobile, and desktop interfaces.
Find out more → https://t.co/DZyfe7aIHd https://t.co/z4xAKAtcah· 34d@generalcatalyst"We are as focused on spatial-temporal dynamics as Anthropic was on code."
@gen_intuition's @PimDeWitte explains why the next frontier in AI is teaching models how the physical world unfolds in time. These world models unlock robotics, simulation, and scientific discovery.
Congratulations to Pim and the team on their Series A. We’re proud to keep backing them.
Revisit @max_rimpel's full conversation with Pim below.· 34d@GreylockVCStudents: Applications for Greylock Techfair 2026 are open.
Sneak peek at this year's lineup:
@cognition, @tryramp, @baseten, @braintrust, @glean, @ListenLabs, @harvey, @nuro, @llama_index, @SnorkelAI, @cogent_security, @Netic_AI, @opal_sec, and many more.
Greylock Techfair is an invite-only recruiting and networking event for top technical students. Meet founders, engineering leaders, and hiring teams from 40+ startups. Explore internships and full-time roles. Get an early look at the next generation of category-defining companies.
Since 2014, past attendees have gone on to join companies they first discovered at Greylock Techfair, including @stripe, @figma, @databricks, and @discord.
👉 Apply by July 12th: https://t.co/sOoF9JNTAG
Spots are limited. Rolling approvals until capacity is met.· 34d@generalcatalyst"The fintechs wanted to remove the friction, and hence remove the advisor. But that friction was passed down to the end client."
For over a decade, the promise was that tech would replace the financial advisor. Instead, all the complexity and anxiety got handed to the customer. @rronrex founded @arca_wealth to fix that. AI handles the back-office work so advisors can focus on clients and help them reach their full financial potential.
We're proud to lead Arca's Series A.
Congratulations to Rron and the entire Arca team.
Watch @Natlevywesthead's full conversation with Rron.
Chapters
00:00 — Where the Trust Began
01:21 — Kosovo to America
02:47 — Lessons from Cadre and Plaid
05:25 — It's Not About the Money
09:16 — The Two Decisions That Change Everything
13:22 — The Millions Nobody Serves
15:48 — Inside Arca
17:25 — Why AI Shouldn't Replace Your Advisor
19:54 — The Cockpit Every Advisor Wants
23:27 — What the AI Does Behind the Scenes
29:54 — AI Services vs. AI Software
34:38 — What a Restaurant Taught a Fintech
38:59 — Hiring People Who'll Outgrow You
51:41 — Becoming Your Financial Everything· 34d@perplexity_aiIntroducing Computer for Counsel.
Computer now connects the research databases, document tools, and matter-management systems lawyers use every day. Pull citable sources from @midpageAI, @LegalZoom, @Docusign, @netdocuments, and more.
Available for all Pro and Max subscribers. https://t.co/El3028Ua7P· 35d@GreylockVCWe are thrilled to share that @glennkelman has joined Greylock as an Executive in Residence.
Glenn co-founded Plumtree Software, which went public in 2002, and spent nearly two decades as CEO of @Redfin, building it into the leading technology-powered real estate brokerage.
At Greylock, he'll work directly with founders on leadership, company building, and the hard parts of scaling that don't fit neatly into a board deck. We know from working alongside entrepreneurs that building a great product is only part of the journey. The personal growth required to lead a company through every stage of scaling matters just as much.
Early in his time at Redfin, Glenn was under enormous pressure and trying to do it all alone. A Greylock partner played a formative role in helping him find a different way to lead. In his own words:
"A Greylock partner named James Slavet took me on a walk along a strip of airport hotels by the Bay. He made me feel that I was supported, but also that I could change. After that walk, I told our board what was really wrong with our business, so we could work together on how to fix it. I began trying to become a better colleague to all the beautiful Redfin people who made me successful.
If I could go on a walk like that with any of you, my work here would be done."
Welcome, Glenn!· 35d@GoogleDeepMindWhat happens when millions of AI agents start negotiating, transacting, and delegating to one another?
@weballergy joined our podcast with @fryrsquared to explore the rise of agentic economies – and how we can diversify agent decision-making to avoid AI groupthink.
Timecodes:
00:00 Intro
1:07 Defining AI agents
4:44 Agentic exploration in science and research
15:46 Delegation between agents
22:46 Agentic security and traps
29:31 Building an agentic economy
33:22 Cognitive monoculture
36:29 Distributed intelligence· 35d@accelContext changes everything in speech recognition. Big congrats to the AssemblyAI team on this one.· 36d@MistralAIIntroducing Mistral OCR 4. It creates structure with bounding boxes, block classification, and inline confidence scores in 170 languages. 🧵👇 https://t.co/jR78NkL4xK· 36d@MistralAIMistral officially has 1,000 team members around the world!
Thank you to our incredible talent for believing in the mission and contributing to what we are building each and every day.
You too can be a part of the next chapter in our journey as we continue to grow and serve our customers. Apply for open roles at Mistral across all functions here: https://t.co/tFpS6Fcjg9· 37d@sequoia12 years ago a top scientist said AI would wipe out radiology.
Computer vision spread through every scan – and radiology demand went UP.
Hospitals hired more radiologists.
@Nvidia’s Jensen Huang on why we keep confusing a job's tasks with its purpose. https://t.co/NWhGQcAHQi· 41d@traestephensHe's baaaaaack https://t.co/do65FWakEi· 41d@astrangeThe story of how @SantiaSua is building Addi is a masterclass in scaling an AI first company. 1 in 4 Colombians now uses Addi to shop, pay, and soon to bank. Addi has 200+ agents in production - they counterintuitively started in legal! - for full automated legal, onboarding & CS.
For anyone thinking about hiring & scaling a global workforce & building agent, checkout the full conversation with me & @GEVS94 @Addi_Col· 41d@julesyooShared the AHIP 2026 stage last week with @aboehler and @peter_orszag to debate healthcare affordability + policy. Some of our takes:
🔥 Healthcare affordability has not improved in the last few years, and has even gotten worse in several areas.
🔥 We've also all aged ~20 years in the last 3 due to the rate of policy and regulatory change in gov't health insurance markets - e.g. MA v28 + new rate notice, Medicaid work requirements under HR-1, ACA subsidy cliff, etc... the health plans are fatigued (and still have a ton of work to do to actually implement and execute against these changes).
🔥 And at the same time - innovation is accelerating! From GLP-1s, cell and gene therapies, to AI-enabled care models. While these technological advancements promise profound health impact, broad affordability for them has not yet been solved. Good policy needs to address market failures related to risk pools, time horizons for underwriting said risk, and pragmatic payment models with teeth behind them.
🔥 Consumer-directed cash-pay is still a rounding error of total spend today... but we should keep an eye on the trend: consumers are waking up to the fact that they're burning $1,000s inside their own deductible for opaque, overpriced services + a miserable UX. Cash-pay-optimized suppliers are rushing into that void - and people are voting with those deductible dollars. e.g. @function, @CounselHealth, @truemed, et al are winning a rapidly growing following based on superior value and experience that the insurance-optimized system is not incentivized to deliver. Will these kinds of players eventually constitute the future version of health insurance?
🔥 In 2025, providers were out front on AI while health plans were playing defense... but in 2026, the plans are clearly leaning in, across both internal ops and external member- and provider- facing experience.
In 5 years, if we're saying healthcare affordability got better, it'll be because we finally embraced AI to bring down the cost structure of care on BOTH ends of the acuity curve. The low end is obvious. The high end - the sickest, most comorbid patients - is where we were most hesitant to say exactly what the near-term impact could be.... but therein is the grand challenge for builders: bend the cost curve in areas where the system looks the most AI-proof.· 42d@sequoiaDon't mistake task for purpose.
@Nvidia's Jensen Huang in conversation with @Konstantine: "Typing is not the job of a software engineer. Coding is not their job. Solving problems is their job."
Jensen says AI is creating more jobs, not destroying them. Why? AI lets us solve problems better than ever.· 42d@julesyoomessi - still got it. 🎩🪄· 43d@traestephenshttps://t.co/RoirVUKP3z· 44d@traestephensWow. Exciting day of soccer. https://t.co/1NEZTKJXfT· 44d@FoundersfundWill the Mafia win? Final game of the night on Thursday, 6/18. https://t.co/3nrsOh5juC· 44d@Foundersfundhttps://t.co/FpvYCd7jRV· 47d@fintechjunkieJust One More Hour
It's 11pm and you've already put in a full day. You know you should shut the laptop but there's a feature you've been thinking about since dinner. And you know that if you work with Claude Code one more hour it will exist before you go to bed.
So you stay.
Gamblers say "one more hand." Bingers say "one more episode." And now Builders everywhere are saying "Just One More Hour." A new addiction is sweeping through the Startup ecosystem.
THE MACHINE CHANGED
For all of human history, meaningful work took time. Projects unfolded over weeks, months and years. Progress was measured in milestones that were spaced far enough apart that shutting down for the night was the obviously rational choice. Nothing important was going to happen between 10pm and 8am, so why stay at the desk? Sleep was important and the work would still be there in the morning.
That math just broke.
One extra hour with Claude Code or Codex and your project is materially farther along. A few extra hours and your to-do list visibly shrinks. Features materialize and bugs get eradicated. Testing happens so quickly that you can build loops that ship, test, fix and ship again while you watch. Websites and mobile apps appear almost out of thin air. For the first time in history, a single hour of work produces SIGNIFICANT output you can see, touch and demo.
And the human brain is wired to respond to this kind of stimulus. Visible progress and accomplishment trigger dopamine. The gap between effort and reward has collapsed from months to minutes, and every Builder's reward circuitry is lighting up like a slot machine that actually pays out.
Which leads us to a question that needs to be answered: Is this dopamine earned or unearned?
EARNED VS. UNEARNED DOPAMINE
The release of dopamine creates a reward circuit in the brain that equates specific experiences with pleasure and reinforces them via intense cravings.
Cheap sources of unearned dopamine train your system to want more cheap sources of dopamine. Post on social media to chase “likes” and you get a dose of unearned dopamine. Become an “angry keyboard warrior” and the crowd’s retaliation feeds you unearned dopamine. Day trade stocks and watch the minute-to-minute fluctuation in price and you’re swimming in unearned dopamine.
Sources of unearned dopamine are everywhere. Your brain gets paid and your life gets nothing.
Earned dopamine is different. It's the reward for actual accomplishment: The deal you’ve been working on for months that closes, the marathon that you finish after a season of training, or the product that your team has been working on for months finally ships. Nobody writes thought pieces warning you about the dangers of actually finishing things. Earned dopamine is supposed to be good!
But here's an uncomfortable new truth:
THE DOPAMINE FROM JUST ONE MORE HOUR IS EARNED!
The progress is real, the feature you shipped actually works and the bug you’ve been chasing down is actually dead. This isn't a casino trick where they trick you into thinking you’re winning. You ARE winning.
And that's exactly what makes this a dangerous addiction the Startup community is facing. Nobody intervenes on an addiction that produces results. Nobody’s going to tell you to stop being productive.
COMPRESSION IS WHAT TURNS A SUBSTANCE INTO A DRUG
I try to come up with analogues to explain new conceptual frameworks, and what follows is the best I’ve come up with. It’s not great, but it’s useful in making a profound point.
The analogue: People chewed coca leaves for thousands of years without a societal crisis and the molecule was never the problem. We learned to concentrate it and speed up its delivery and the harmless substance suddenly became cocaine. Concentration plus delivery speed can turn a benign substance into a drug.
Builders have been getting high on shipping forever. But the shipping high used to come a few times a year after months of grinding, with long stretches of friction in between. AI coding tools took the same substance and removed the friction and sped everything up. The high is now available hourly, on demand, with no natural cooldown.
What nobody wants to internalize is that the friction was protective. The sheer act of typing to code, compile times and waiting on your teammates weren't useless inefficiencies. They acted as natural guardrails and forced rest into the system whether you wanted it or not. AI just removed all of them at once and we’ve embraced it without any understanding of the consequences.
THE GAME ON THE FIELD
If this were purely a personal willpower problem, the advice would be easy: Log off, touch grass and get some sleep. But the Startup ecosystem has a nasty structural twist that turns a private temptation into a collective trap.
Speed has become the new moat.
When every team has access to the same models, the same tools and the same compute, it’s tempting to believe that the only differentiator becomes how many hours you feed the machine. Your competitor's product is materially better every single morning which means shutting down at 9pm is no longer a lifestyle decision. It's a competitive decision.
I've written before about how people aren't irrational, they're rational actors responding to the game that's on the field. So take a look at the game that's now on the field for Founders: The cost of an extra hour is low, the output of an extra hour is visible, and everyone else is putting in the hour. Run that math and "Just One More Hour" isn't an addiction at all. It's the dominant strategy.
That's how arms races work. Every individual decision is rational and the collective outcome is brutal.
THE TRADE YOU'RE REALLY MAKING
So is the conclusion "work less"? No. The conclusion is that many Builders could be mispricing the trade.
When speed becomes abundant, it stops being the edge. The machine supplies the speed now. What the machine cannot supply is the judgment that turns it into velocity: Knowing what to build, what to kill, what actually matters to a customer, and what's a shiny object wearing a customer costume. Speed can drive you 100 miles an hour in circles. Speed PLUS Judgment gets you the Velocity to push your business forward. An opinionated perspective about the solution is what’s scarce in an AI-saturated Startup ecosystem, and it’s precisely this clarity of thinking that degrades for most people after midnight.
Anyone who has reviewed their own 1am decisions in the morning knows this.
A well-rested team pointing the machine at the right problems will beat an exhausted team pointing it at the wrong one, and the gap will compound every week because the machine multiplies the quality of its instructions.
THE MACHINE SUPPLIES THE SPEED. YOU SUPPLY THE JUDGMENT THAT TURNS INTO VELOCITY.
Act accordingly.
ONE MORE THING
I want to be honest about where this essay comes from. I'm back in the Builder's seat, and I live in Claude Code 10+ hours a day right now. I've always been a hard worker and I'd put my career track record of outworking the competition up against just about anyone's. I thought I understood what the pull of work felt like but this is different.
The pull of "just one more hour" is showing up for me every single night without fail, and it arrives wearing the most convincing disguise an addiction has ever worn: Genuine, demonstrable, earned progress.
I know the dopamine is real and I’m not ignorant of its long term impact. I’m very self-aware. But what I do know is that tonight, somewhere around 11pm, I'll feel the pull and work a few more hours anyway.· 48d@GreylockVCGreylock is excited to lead Niteshift’s $7M seed round, alongside @AmplifyPartners, @BoxGroup, and @svangel.
Coding agents are quickly becoming capable of taking on longer, more complex engineering tasks. But the bottleneck has shifted from what agents can write to whether they can run, test, and verify their work in real development environments.
Introducing Niteshift, the full-stack cloud for coding agents.
Niteshift lets teams run frontier and open-source coding agents inside fully configured cloud dev environments, so agents can return pull requests with evidence that they work.
@smehmood and Conor's infrastructure instincts come from years spent building at Datadog, where they worked across complex codebases, cloud development, observability, and production environments. They understand that the hard part is not just writing code, but giving software the environment, tools, and feedback loops it needs to work reliably.
We’re proud to partner with Sajid, Conor, and the Niteshift team as they build the cloud for coding agents. Read more on the Greylock blog and in TechCrunch.· 49d@julesyooI always think about this in the context of the AI-native “medical record” - and would include AI chat transcripts in there as well. Between ambient recordings + chat transcripts (effectively a form of PRO), AI care / AI care management companies will have far more health context on patients than traditional EHRs. Trad EHR notes are inherently abstractions optimized for billing versus true narratives of a) a person’s health situation and b) the clinician’s reasoning about how they decided the diagnosis and course of action.· 49d@astrangeClaude is telling me not to procrastinate https://t.co/y0PoqpWSKP· 50d@astrangeEven the best jogging strollers are hard to push and muck with your stride 🏃♀️ maybe this is the solve 😂· 52d@FoundersfundWho's lying? https://t.co/3zs09HYgeM· 54d@MistralAIWe're taking on the hardest problems in the real world 🏗️🚚 🛫⚛️
Today at The AI Now Summit, held at the Louvre, we announced AI solutions for aerospace, automotive, energy, and physics.
Deployed in production at @Airbus , @BMW, @EDFofficiel , and more.
More below: https://t.co/L6Vq7RdABg· 62d@fintechjunkieThe Score Isn't the Game
The AI personal finance race is on. OpenAI can access your bank feed with permission. It’s a great start but nowhere near enough.
I love seeing innovation, progress, and the democratization of intelligent advice in personal finance. ChatGPT and Anthropic are making waves which shouldn’t be a surprise to anyone paying attention. The category needs more thoughtful capital, more product muscle, and more shots on goal. It’s too important to ignore.
But personal finance isn’t just about budgeting or tax planning. Society has shifted in a profoundly negative way over the past decade such that financial nihilism is more the norm than the exception. People don’t feel like they can successfully navigate today’s expensive world and many people believe it’s getting worse, not better. So let me say the uncomfortable thing out loud:
Connecting your accounts isn't going to fix this.
Not even close.
And waiting for the government to fix it for you isn’t a strategy.
It’s hopium.
The Dashboard Trap
Giving credit where credit is due, the OpenAI / Hiro / Plaid integration is a real step forward. Seeing all your money in one place is genuinely better than logging into seven different apps and trying to keep a mental running total. The industry has been working on this problem since Mint launched in 2007, and a number of companies like Monarch, Origin Finance and YNAB are fantastic tools that millions of users love and use every day. ChatGPT will open up this functionality to tens of millions of new people over time which should be applauded.
But seeing your financial chaos isn't the same as fixing it. It's the equivalent of stepping on a smart scale every morning. The scale will give you useful data but it doesn't make you healthier.
Aggregation tells you the score of the game you're playing but it doesn't change the game.
The Rule Followers Are Losing
Millions of people did everything they were told to do by getting the degree, taking the stable job, saving and avoiding dumb debt. They followed the playbook their parents handed them and ticked all the boxes society told them to.
These same people are frustrated that their effort hasn’t given them the prize they thought they were chasing.
Housing prices rose 48% between 2019 and 2024. Median income rose 22%. Childcare runs 3x what HHS considers "affordable." Student loan repayment averages 20+ years and puts families behind on every major life event. The median 401(k) balance for Americans approaching retirement is a meager $65,000.
These are just facts. The numbers represent how much the social contract has changed.
I've been writing about this for a while and coined the phrase P(win) = 0 which is meant to represent the awakening people have when they believe the probability of winning is zero if they continue to follow the rules as they’ve been laid out. When they realize the math no longer maths because life became expensive and people’s incomes didn’t keep up. The conventional path stopped producing conventional outcomes a long time ago and people are frustrated that everyone keeps insisting the conventional path is the answer.
When P(win) = 0, any alternative becomes rational, even the ones that seem dumb on the surface.
It’s why 55% of Gen Z investors prefer crypto to the traditional financial ecosystem. It's why Kalshi and Polymarket went from fringe to mainstream in eighteen months. It's why 0DTE options trading has exploded. It's why "save and splurge" lives in the same wallet at the same time. People aren't being irrational. They're being rational because they feel stuck inside a game they've concluded they can't win.
Better Dashboard, Same Game
Now connect the dots with the new “connect your financial accounts” announcements.
If the underlying problem is that conventional financial advice no longer produces conventional outcomes for huge swaths of society, then aggregating bank feeds and pointing a large language model at the resulting dashboard gives you faster, prettier, more personalized versions of the same advice that already isn't working.
Calling that democratized intelligent advice is generous. It's democratized access to a playbook that's convenient but outdated. It’s great directional advice that doesn’t actually help people address root cause issues. It’s a pain killer that makes you temporarily feel better.
But the truth is harsh. A better visualization of a losing position is still a losing position.
The Real Moat
I've written before that in a world where anything can be built in a weekend, the last real moat is an opinionated perspective on the solution. I believe this is more true in the personal finance category than almost any other.
Aggregation will commoditize and so will the model layer. The edge is going to live somewhere else entirely.
It's going to live with whoever has a genuine point of view on what's actually wrong and a framework for what to do about it that respects today's realities rather than the realities of 1990.
That means saying out loud that the old playbook ("save 10%, retire at 65, 4% rule") is structurally broken for a huge portion of the country. It means building an advice layer that takes today’s reality as the starting point:
The structural gap between income and cost of living is real and can't be budgeted away
People need to live the lives they want to live now, not just plan for a hypothetical future
The goal isn't perfect optimization. The goal is making the game winnable again.
That's a product position. Most of what's being built right now is a feature position. There's a real difference.
Building It
Which brings me to what I haven't really said publicly yet.
I'm building it.
I've spent the last year quietly putting together what I think the answer looks like. At its core, it’s an AI financial workspace built on a deeply opinionated framework I've been refining that starts with today’s reality. Everything I've been writing about for years is baked into the architecture.
It's a fiduciary-minded decision engine that takes a position on what's actually broken and gives people specific, actionable moves that account for the world as it actually is, not the world we were promised.
More to come when it's ready (but we are hiring!).
I'm genuinely glad the big AI labs are wading into personal finance because the category has been underserved for decades and the work being done now will move the entire space forward. Real competition is good for users.
But the people who've been doing everything right and still aren't winning don't need another dashboard.
They need someone willing to say:
You're not crazy. The social contract was rewritten so here's how to play the game and win.
That's what I'm building. Wish me luck.
Onwards and upwards,
Fintechjunkie· 72d@natfriedmanStarted work at Meta this week. My job is to make amazing AI products that billions of people love to use.
It won't happen overnight, but a few days in, I'm feeling confident that great things are ahead.· 392d@natfriedmanCivilization is constant maintenance· 422d@natfriedmanUSA: passing 1000 state bills to slow down AI
China: passing american AI on leaderboards· 426d@Shaun_MaguireOn the "birtherday" day of @realDonaldTrump ,
Let's celebrate the best President ever - Barrack Obama... https://t.co/rNJXKlH25j· 2236d@chamathhttps://t.co/baZ2XV4BHI· 27d@JasonMy team at The Syndicate is meeting with family offices to review some very interesting deals we're working on.
If you would like to take a meeting, please email fo@launch.co· 27d@JasonThe appearance of impropriety that the 47th administration is signaling is the perfect way to accelerate the democratic socialist movement, the America First/America Only movement & the creation of @TuckerCarlson's 3rd party.
We need to get money out of politics by paying the President a top salary, say the median/mean salary of the top publicly traded companies ($25M a year/$100m a term).
We then need to do something similar (at a smaller scale) for VPs and members of Congress.
Finally, we have to limit all donations and cap SuperPACs' spending.
Our system is completely bought and sold at this point.
Please explain to me why you think I'm wrong.· 27d@Jason⚔️ @jalenbrunson1 ⚔️ https://t.co/3kxGMQgJ9k· 27d@eladgilSource of question -
https://t.co/9BRcsFknOL· 27d@xaiGrok Build is now installed in Railway sandboxes· 27d@HarryStebbingsThe single best follow in tech on X right now is @nikesharora.
One of the best operators of our time spewing wisdom on a daily basis.· 27d@KTmBoyle“This man seemed enthusiastic and ambitious, but I must confess to holding some serious doubts about the dramatic scope of his plans and his ability to pull them off… well, he did the thing.”
This is the story arc. Aspire to this. 🇺🇸🚀💪· 27d@PalmerLuckeyBe careful with your fireworks this weekend!· 27d@saranormoussuper impressive example of combining machines-asking-humans in order to efficiently direct use of compute against a goal· 27d@chamathyup.· 27d@packyMThis is real btw https://t.co/WOyE5WLijL· 27d@zebulgarSince everyone is sharing interesting World Cup stats
My interesting one, is that so far, every winning team
Has been the one that has scored the most goals in their game· 27d@packyMThe next 250 years of the American project should be unimaginably abundant, if we can keep it.
Read Scott Nolan's vision for America's Next 250 in today's Not Boring. https://t.co/FzdY5ySngJ· 27d@HarryStebbingsDario just declared war on open-source.
Anthropic's message is clear: open source could destroy the entire AI business model, and Chinese open-source models are the cause.
I sat down with @jasonlk & @rodriscoll to discuss it, along with the biggest news in tech this week:
- Anthropic & Dario Declare War on Open-Source
- Coinbase Slashes AI Spend 50%: Is the Token Bubble Bursting?
- Kalshi's $40BN Valuation & Impending IPO
- Bending Spoons: The Smartest IPO of 2026 & the $100BN SaaS Roll-Up Play
My notes below:
1. Anthropic Is Laying the Foundation for a Deal With the U.S. Government on Chinese Models
Anthropic accused Chinese open-source AI competitors of “brazen theft” through model distillation that violates its terms of service. Rory noted the hypocrisy, given that U.S. frontier models originally scraped external IP. Still, Anthropic appears to be laying the groundwork for a regulatory trade: complying with domestic access restrictions in exchange for a federal ban on distilled Chinese models.
2. Jason Lemkin’s Response to Brian Armstrong’s AI Tweet
Jason dismissed Coinbase CEO Brian Armstrong’s 50% LLM cost-cutting post as “performative social media,” arguing that savings matter little if core revenue is flat or shrinking. He believes AI must actively drive top-line growth. Rory defended the move as “cost management 101,” saying cash-conscious enterprise executives will quickly emulate it to curb runaway frontier model fees.
3. CEOs Are Struggling to See the ROI From AI
Massive enterprise spending on AI tokens is failing to deliver the expected revenue or productivity gains, leaving CFOs searching for measurable operational lift. Jason noted that adding millions in AI spend can still produce the same growth rates as prior quarters. Rory argued that AI spending must clearly accelerate software delivery or create definitive bottom-line savings as boards push back on reckless “token maxing.”
4. We Are All So Aligned in Wanting AI to Win
Jason warned that the U.S. economy is structurally addicted to AI, with 40% of the S&P 500 tied to the boom, making society eager to prop it up to protect 401(k) portfolios. Rory countered that protectionism artificially inflates intelligence costs for the broader economy. He compared it to banning IBM PC clones in the 1980s just to protect IBM’s stock price, calling the blocking of low-cost open-source alternatives “fricking dumb.”
5. Bending Spoons and the New Playbook for B2B Revenue Arbitrage
Bending Spoons’ $20 billion public valuation marks a shift toward tech roll-ups that drive profitability through price hikes and cost-cutting rather than organic user growth. Jason predicts this playbook will expand into mature B2B SaaS. By acquiring sticky but underperforming platforms like Marketo, Asana, or PagerDuty and injecting hungry talent, operators can rapidly improve retention and capture massive revenue arbitrage.
(links below)· 27d@garrytanPlease oppose ACA-7
If you live in California take a moment to click the link here and tell your representatives
We want an end to racism of all kinds in our schools· 27d@chamathPart of what 8090 will be able to show, over time, is that companies powered by us generate return on capital employed (ROCE) that exceed their peers.
We help companies do this by encoding their "edge" into software that can recursively improve vs leaving it as tribal knowledge or tacit conventions that sit in people's heads but nowhere else. Because when those who know what to do leave a company, it creates a hole and those that remain often find themselves treading water or moving backwards as a result.
From a recent BCG report:
"As the cost of capital has returned to long-term averages (8% to 11%), roughly half of large US companies cannot deliver returns that exceed that.
Persistently low returns on capital (as measured by return on capital employed, ROCE) are surprisingly common across public companies, affecting about one in seven companies globally.
These underperforming businesses represent a widespread challenge confronting many portfolios. Even healthy businesses grapple with divisions that underperform, dragging down overall profitability and complicating strategic decision-making."· 27d@altcapTake me home! 🇺🇸🇺🇸🇺🇸 https://t.co/aWWYyI7Isr· 27d@packyMIf I were FIFA, I’d think very carefully about what happens to haughty Europeans who try to impose rules we don’t like on us without our consent on our home soil around July 4th. https://t.co/4KQYnQ5htZ· 27d@pmarcaInteresting.· 27d@pmarcaInteresting.· 27d@pmarcaMany such cases.· 27d@levieIf you’ve ever wondered why we will need 100X more AI inference in the future, and what it’s going to be driven by, this is another good example.
Devin pushes forward an idea of agentic mapreduce, which means we’ll now have swarms of agents that are processing large amounts of data (code) to handle tasks that humans never could have done before.
“Devin maps relevant signals across the repo, fans out focused agents over bounded shards, reduces their findings into one report, then verifies serious vulnerabilities in isolated sandboxes before marking them confirmed.”
In this case it’s code security, but there are tons of other use-cases in code and knowledge work. We see this at Box with customers that want to process and understand millions of documents for risk, insights, relationships, and more. This will play out in pharma, banking, and many other industries across all forms of unstructured data.
As an aside, these types of capabilities are generally only possible when you can deploy a variety of models (both the frontier and lower cost) because of the sheer amount of tokens that go into these use-cases. This is going to be a major value proposition for the applied AI layer.· 28d@zebulgarAMERICAAAAA FUCK YEAHHHH
2-0
🇺🇸🇺🇸🇺🇸 https://t.co/oyjppKLLf7· 28d@garrytanhttps://t.co/gTrjV7mcxf· 28d@martin_casado“In the near term, some routine tasks like coding and debugging will fall back to Opus 4.8” 😳🫠· 28d@saranormousnuclear powered GPUs!!· 28d@perplexity_aiClaude Fable 5 is once again available in Computer as an orchestrator model. https://t.co/U65gnvja30· 28d@altcapThe American Dream has never belonged to one party. It belongs to every child. Grateful to be in Newark w @SenBooker signing up kids for @TrumpAccounts. In NJ over 1 M kids have at least $250 waiting in their free, lifetime investment account. 🇺🇸🚀 https://t.co/QIPzh7f0aw· 28d@martin_casadoIt's baaaaaaaaaccckkkk https://t.co/jP4sk1A8zA· 28d@garrytanMost underreported story in all of tech· 28d@Lightspeedvp.@bsomaia on AI valuations:
“Are we in a super cycle? Yes.”
“Are we in a very rich valuation environment? Yes.”
“Will there be investments that despite this valuation environment will be extremely rewarding? Yes.”
"Will there be investments that turn out to lose money? Yes."
"All of those things can be true."
From Startup Policy Forum with @ShwetaRKohli.· 28d@cohereHappy Canada Day!! 🇨🇦
We're so proud to be founded and growing in Canada. Let's keep building💡· 28d@bgurleyWeChat has been a little quiet on AI front & perhaps even falling into being considered a laggard. But this week, Pony Ma presented a new vision on how to turn "mini-apps" (still no real equivalent in US) into AI handshakes. Manny partners announced. Could be a powerful move. Read more below.· 28d@LightspeedvpThe team at @unconvAI has released their first model, Un-0!
Un-0 is an image-generation model powered by a simulated system of coupled oscillators.
Instead of the massive arithmetic operations that power most AI, the model uses simulated physics. Thousands of virtual oscillators influence each other, sync up, and settle into patterns. Feed it an input, let the oscillators interact, and the resulting pattern is the output.
On ImageNet 64×64, Un-0 achieves an FID score of 6.74, matching the quality of leading conventional image generation methods when they were first published.
The company has stated that, to their knowledge, this is the most capable image generator to date that uses a simulation of a physical dynamical system.
Unconventional AI believes energy efficiency demands a fundamentally different computer, one where physics does the computing. We believe this is a meaningful milestone in their goals of achieving ~1,000x energy-efficiency gains in modern AI.
Congratulations to @NaveengRao, @MeeLan_Lee, @sachour__, @MCarbin, and the entire Unconventional AI team. We are incredibly excited for what’s ahead.
CC: @guruchahal, @ravi_lsvp, and @buckymoore· 28d@xaiIntroducing Voice Agent Builder: a no-code platform to create human-like voice agents with Grok Voice.
Available today at $0.05 / min.
https://t.co/kUkF7zqvfR https://t.co/OCIq1oDYar· 28d@LightspeedvpA big congratulations to the @Arena team on hitting $100M ARR!
This milestone announcement comes just eight months after the launch of their evaluation product.
Arena hosts millions of users who can work in real time with AI agents, testing and comparing different AI models. One primary feature is their AI model performance leaderboard, which has been generated from over 10 million user evaluations.
Incredible milestone @ml_angelopoulos, @infwinston, @istoica05, and team. We’re proud to see Arena continuing to develop and advance the frontier of AI through community-driven evaluation.
Learn more in the post below.· 28d@eladgil🔥 @harvey on fire· 28d@accelOur 2026 Founders Factory report is now live!
Europe and Israel's startup ecosystem has entered a new phase. Employees from 322 VC-backed unicorns have gone on to found more than 2,300 startups across the region, raising over $42B and creating a self-sustaining ecosystem where talent, experience and capital compound with each generation of founders.
At the same time, AI is accelerating this momentum, creating a new generation of companies that are scaling from zero to unicorn status in record time.
In this year's report, produced in partnership with @dealroomco and @RevelioLabs, we explore:
➡️ How more than 2,300 startups have emerged from Europe and Israel's unicorn founder factories
➡️ How AI is creating Europe’s fastest-ever unicorns, with 1 in 5 new unicorns since 2023 reaching a $1B valuation in under 2 years
➡️ Which European startup ecosystems are creating the most founder factories, with London producing more than any other city while Sweden’s founder factories generate nearly 3 times more startups per unicorn than the European average
➡️ How the archetype of Europe’s newest unicorn founders is changing, with today’s founders more likely to come from Big Tech and academia
Read the full report here ⬇️ https://t.co/sqAjYX47CD· 28d@cohereThe future is yours to build.
Cohere x @AstonMartinF1 https://t.co/K2P5iJJ6ue· 28d@KTmBoyleEndless summer. https://t.co/vXtZeWSGww· 28d@altcapThank u to Sec Lutnick, @NotTomBrown & @WhiteHouse for moving so quickly to get America’s frontier models back online. Finding the safety balance was never going to be easy - but America is winning & this rapid response & new framework will accelerate us yet again. 🇺🇸🚀· 28d@levieThings seem to be ending up in a better spot with Fable, and presumably GPT-5.6 next. What we have now is the initial precedent for what frontier model releases (or at least those that have significant coding and cyber capabilities) could look like going forward. This would presumably apply to bio and other categories of risk that have been identified by AI safety groups.
From the Anthropic post:
“3. A shared industry framework. Although we have reached a constructive resolution, these events have made clear that the industry needs a consistent way to assess and fix potential “jailbreaks” of AI models (techniques that bypass a model’s safeguards).2 A shared standard for judging the severity of a given jailbreak would help AI developers triage new findings as they arise, launch highly-capable models with greater safety, and communicate the level of risk consistently to government and industry partners. Together with Amazon, Microsoft, Google, and other Glasswing partners, we’ve started to develop such a framework, and we outline it below.
4. Deeper government collaboration. We’re also strengthening our level of collaboration with the US government on new pre-release testing, information sharing, and research collaboration. We describe this deeper collaboration in the final section.”
It’s been a messy process to get here, but at least there’s some semblance of a framework that could be practical. The only note of caution here would be that there’s a lot of subjectivity that goes into various risks and their actual levels of exploitability in practice. We’re likely going to be living with a framework that requires heavy judgment and back and forth between labs and the government for major releases.
The best we can hope for is that this is a relatively efficient process, and hopefully has ways of being sped up for incremental version updates in models. It would be a bad outcome if every release after this level of threshold of capability required the same review process, and we don’t get the same rate of breakthroughs we’ve been seeing.· 29d@AnthropicAIClaude Fable 5 will be available again globally tomorrow.
After a series of productive conversations with the US government, we're redeploying the model with a new set of classifiers to target and block more cybersecurity tasks. In the near term, some routine tasks like coding and debugging will fall back to Opus 4.8. We’ll continue to refine these classifiers over the coming weeks to reduce false positives and better distinguish genuine misuse from legitimate requests.
We’ve also begun drafting a consensus framework—with Amazon, Microsoft, Google, and other Glasswing partners—for assessing the severity of AI jailbreaks and how AI developers should respond to them. We invite other industry partners and model providers to join us in this effort.
Finally, we’re scaling up our collaboration with the US government on model testing and safeguards. This will include pre-release access to models and safeguards for evaluation, information sharing on jailbreaks and misuse, and dedicated resources for joint research.
Thank you to our users for your patience, and to our partners across the government, industry, and the research community who worked alongside us to make Fable 5 available again.
Read our full blog: https://t.co/VHyum831ri· 29d@bgurleyThis is absolutely correct. The only way to fight "open" is with "open." Glad to hear a voice in Washington that understands the real stakes.· 29d@AnthropicAIWe’ve received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5.
We'll begin restoring access tomorrow, and will share an update soon.
We’re grateful to our users for their patience, and to everyone who worked with us on redeploying the models.· 29d@HarryStebbings"After food, energy sovereignty is one of the most important strategic priorities for any nation.
History shows that wars are often fought over access to essential resources, and today, energy is at the center of that competition.
The more countries can produce reliable energy domestically, the more resilient, secure, and economically independent they'll become."
What is more important energy sovereignty or model sovereignty @jameswise @vkhosla @alanchanguk @wolfejosh @traestephens· 29d@martin_casadoHey book lovers.
I'm re-reading Hesse's Steppenwolf, and it's an entirely different book than when I read it in my 20s. The perspective of age changes everything.
What other books are like this? Worth a re-read a few decades later to get a wholey new perspective on?· 29d@cohereWe build differently than other tech firms.
No superintelligence. No competition to spend the most money. We're creating empowering, efficient AI to enhance human potential, not replace it.
With high standards for engineering & innovation, we're proud to support @AstonMartinF1. https://t.co/q5ZoRNV9Qe· 29d@levieWe've been running Anthropic's Claude Sonnet 5 through the Box AI Complex Work Eval, our agentic benchmark that puts models through real enterprise document work end-to-end.
Sonnet 5 holds frontier-class quality on complex multi-step work and pulls ahead of Sonnet 4.6 in several core enterprise domains like Energy (+4.7pp), Retail (+4.4pp), and Professional Services (+2.6pp), and other spaces where unstructured data is heavily complex.
Here are a few examples of wins compared to Sonnet 4.6 to get a sense of some of the more advanced reasoning capabilities in Sonnet 5:
* Financing due diligence: It computed the company's liquidity and leverage ratios from the raw balance sheet, and caught that the source report's own stated debt-to-equity figure understated the leverage, flagging all three loan covenants as violated, not just the ones the document admitted.
* Overhaul cost analysis: It scoped "total cost" to the company's own KPI definitions, correctly separating out Lost Production Cost because the guidance said to track it separately rather than naively summing every number on the sheet. It also caught and handled a broken reference cell in the spreadsheet.
* SKU revenue analysis: On segmented sales data, it computed each product's contribution against the correct subcategory denominator, sidestepping the easy mistake of dividing by the category total, and flagged why no Pet-category SKU cracked the top 9.
Sonnet 5 will be available in the Box AI Studio shortly for customers to build custom agents with.· 29d@a16zVacation looks very different depending on where you work.
Over the course of the year, Germans hold the OOO crown. But not consistently! Swedes and Italians trade places in July and August, and in August, the French take over. Brazil breaks from the European pattern on summer break length. India books vacation days at the last minute.
And birthdays are popular days off, but only the big ones. Turning 31? Not worth celebrating, apparently.
We teamed up with @deel to look at how startup and tech workers around the world actually take vacation, not just how many days they're offered, but when they take them, how long, and whether they take them at all.
North America is stingier than Europe. Most "vacations" are a single day off. Long weekends spike on Mondays and Fridays in summer.
And no matter how hard the Europeans work to make their summer holidays count, Christmas break wins everywhere. The British are proportionately at their desks in peak summer; the Dutch over Christmas.
Except Armenia, for some reason.
Full piece: https://t.co/UpFUh4KTEF· 29d@perplexity_aiClaude Sonnet 5 is now available in Perplexity for Pro and Max subscribers.
You can also select it as an orchestrator model in Computer. https://t.co/UktzCrUZU6· 29d@saranormousMany parts of venture don’t make any sense to founders, and hanging out with the great but always surprising @stanine, being forced to explain how VC works, reminded me of that· 29d@zebulgarEv and Jack thanked me like 5 times for not overly bashing Benchmark in this
I'm not sure if that means this podcast is good or banal, but hey we had some fun· 29d@OpenAIWe’re introducing GeneBench-Pro, a research-level benchmark for a harder kind of AI progress: how well agents can navigate messy biological data, choose the right analysis path, and make judgment calls that real computational research depends on.
https://t.co/AsilnnSxnE· 29d@GoogleDeepMindWe’re shipping 2 major releases:
🔘 Nano Banana 2 Lite: our fastest and cheapest Gemini Image model
🔘 Gemini Omni Flash: now available via the Gemini API and in @GoogleAIStudio to help developers generate and edit high-quality videos.· 29d@bgurleySuper cool momentum here. Groundbreaking public/private partnerships. And the most effective program for inequality — direct wealth transfer. No unnecessary rake. Great to see new incremental use cases. Love this - business supporting local community.· 29d@generalcatalystWe backed @1001ai_ at seed and are continuing to back the team in their $30M Series A.
1001 builds sovereign AI for the critical infrastructure that moves the physical world: aviation, ports, and energy. It's built and governed locally, so operators stay in control.
Congratulations to @b_abughazaleh, Mohammed Fituri, and the entire team.· 29d@KTmBoyle“The problem isn’t freedom. It’s that we’ve lost the primary institution where virtues are formed— the household in which children form one another.” https://t.co/98j4nWHxUO· 30d@accelThe most important professional conversations and the best ideas rarely happen in front of a screen. They happen in the moments between meetings, over the phone and across tables. Yet many of the tools we rely on to capture them were built for a desk-bound world that doesn't reflect how most people actually work. As a result, follow-ups are forgotten, information stays siloed, and great ideas can slip away.
@Heypocket solves this. Built by co-founders @AkshayNarisetti and @gabrieldymowski, Pocket is a dedicated AI device that helps people stay present. Elegant enough that you don’t have to remember to bring it with you and capable enough that it captures everything you need it to. Joining the Pocket journey as we participate in the Seed round, we’re looking forward to supporting the team as they reimagine the way people reflect, recall and work.
Welcome to the Accel family, Pocket! Read more from Cecilia Wang and Matt Robinson ⬇️ https://t.co/SwJqxGBIPn· 30d@xaiState of the art voice APIs from SpaceXAI, now in the Vercel AI Gateway· 30d@a16zAirbnb, 2008. https://t.co/uBQZL8BEUF· 30d@a16zThe a16z Editorial team’s summer reading list: https://t.co/RxvtUDs2xr https://t.co/lWyb54IyJf· 30d@PalmerLuckeyAt least 1,200 people died of heatstroke in France last Wednesday, mostly elderly people indoors.
Even crazier, about 70,000 Europeans die from the same every year, akin to a bi-weekly 9/11. Eurobros, please vote for leaders who will let you open your windows and install A/C! https://t.co/NxK4irX7GG· 31d@eladgilStatus
https://t.co/e9pZtfyPGr· 32d@fintechjunkieThe Bear was……perfect.· 32d@AnthropicAISince June 12, we’ve been working closely with the US government to restore access to Claude Mythos 5 and Fable 5. Today, the government notified us that Mythos 5, our strongest cybersecurity model, can be redeployed to a set of US organizations that operate and defend critical infrastructure.
We’re restoring access for these organizations quickly, and we’re continuing to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again.· 33d@OpenAIIntroducing a limited preview of GPT-5.6 Sol, our next generation frontier model, as well as GPT-5.6 Terra, a balanced model for efficient, everyday work, and GPT-5.6 Luna, a fast and affordable model for high-volume work.
https://t.co/OoM83SyISN· 33d@PalmerLuckeyJust two days until this incredible vehicle goes up for auction! https://t.co/tzCUYBaHyC· 34d@typesfastTIL the Caesar salad was invented in Tijuana. Mexican food really is the best.· 34d@typesfastIs Rippling going to eat every SaaS category?· 34d@sequoiaToday's AI models train once. We don't work that way. We learn continuously, forget what doesn't matter, and retain what does.
That gap is what @dan_biderman and @realJessyLin are closing at @EngramLab. AI that never stops learning, with memory that lives inside the model instead of bolted on as an afterthought.
In our latest Training Data episode we get into why memory is the next frontier: why the brain forgets on purpose, why RAG is a band-aid, and what becomes possible when a model is always training.
00:00 Introduction
00:59 Always Training Explained
01:51 Beyond Context Windows
03:29 Ngram Product Overview
04:34 Adapters And Training Signals
05:32 Internalize Vs Externalize
06:49 Compute And Token Savings
08:19 Teams First Then Individuals
08:51 Memorization Vs Understanding
12:47 Dreams And Offline Digestion
14:08 Training Beats Curation
15:19 Why Everyone Needs A Model
21:44 Bitter Lesson And Architecture
24:44 RAG Killer And KV Cache
31:38 Future Of Memory And Models· 34d@OpenAIWork at OpenAI is being transformed by agents, in every department.
Across our entire company, people are using Codex to do work that is more complex, longer-running, and increasingly cross-functional.
Our internal usage offers an early look at how agentic tools may reshape work as they become more capable and broadly available.· 34d@typesfastWow JSX just added Santa Monica to Oakland service· 34d@GoogleDeepMindGemini 3.5 Flash now supports native computer use.
This built-in tool lets developers build custom agents that can see and take action across browser, mobile, and desktop interfaces.
Find out more → https://t.co/DZyfe7aIHd https://t.co/z4xAKAtcah· 34d@generalcatalyst"We are as focused on spatial-temporal dynamics as Anthropic was on code."
@gen_intuition's @PimDeWitte explains why the next frontier in AI is teaching models how the physical world unfolds in time. These world models unlock robotics, simulation, and scientific discovery.
Congratulations to Pim and the team on their Series A. We’re proud to keep backing them.
Revisit @max_rimpel's full conversation with Pim below.· 34d@GreylockVCStudents: Applications for Greylock Techfair 2026 are open.
Sneak peek at this year's lineup:
@cognition, @tryramp, @baseten, @braintrust, @glean, @ListenLabs, @harvey, @nuro, @llama_index, @SnorkelAI, @cogent_security, @Netic_AI, @opal_sec, and many more.
Greylock Techfair is an invite-only recruiting and networking event for top technical students. Meet founders, engineering leaders, and hiring teams from 40+ startups. Explore internships and full-time roles. Get an early look at the next generation of category-defining companies.
Since 2014, past attendees have gone on to join companies they first discovered at Greylock Techfair, including @stripe, @figma, @databricks, and @discord.
👉 Apply by July 12th: https://t.co/sOoF9JNTAG
Spots are limited. Rolling approvals until capacity is met.· 34d@generalcatalyst"The fintechs wanted to remove the friction, and hence remove the advisor. But that friction was passed down to the end client."
For over a decade, the promise was that tech would replace the financial advisor. Instead, all the complexity and anxiety got handed to the customer. @rronrex founded @arca_wealth to fix that. AI handles the back-office work so advisors can focus on clients and help them reach their full financial potential.
We're proud to lead Arca's Series A.
Congratulations to Rron and the entire Arca team.
Watch @Natlevywesthead's full conversation with Rron.
Chapters
00:00 — Where the Trust Began
01:21 — Kosovo to America
02:47 — Lessons from Cadre and Plaid
05:25 — It's Not About the Money
09:16 — The Two Decisions That Change Everything
13:22 — The Millions Nobody Serves
15:48 — Inside Arca
17:25 — Why AI Shouldn't Replace Your Advisor
19:54 — The Cockpit Every Advisor Wants
23:27 — What the AI Does Behind the Scenes
29:54 — AI Services vs. AI Software
34:38 — What a Restaurant Taught a Fintech
38:59 — Hiring People Who'll Outgrow You
51:41 — Becoming Your Financial Everything· 34d@perplexity_aiIntroducing Computer for Counsel.
Computer now connects the research databases, document tools, and matter-management systems lawyers use every day. Pull citable sources from @midpageAI, @LegalZoom, @Docusign, @netdocuments, and more.
Available for all Pro and Max subscribers. https://t.co/El3028Ua7P· 35d@GreylockVCWe are thrilled to share that @glennkelman has joined Greylock as an Executive in Residence.
Glenn co-founded Plumtree Software, which went public in 2002, and spent nearly two decades as CEO of @Redfin, building it into the leading technology-powered real estate brokerage.
At Greylock, he'll work directly with founders on leadership, company building, and the hard parts of scaling that don't fit neatly into a board deck. We know from working alongside entrepreneurs that building a great product is only part of the journey. The personal growth required to lead a company through every stage of scaling matters just as much.
Early in his time at Redfin, Glenn was under enormous pressure and trying to do it all alone. A Greylock partner played a formative role in helping him find a different way to lead. In his own words:
"A Greylock partner named James Slavet took me on a walk along a strip of airport hotels by the Bay. He made me feel that I was supported, but also that I could change. After that walk, I told our board what was really wrong with our business, so we could work together on how to fix it. I began trying to become a better colleague to all the beautiful Redfin people who made me successful.
If I could go on a walk like that with any of you, my work here would be done."
Welcome, Glenn!· 35d@GoogleDeepMindWhat happens when millions of AI agents start negotiating, transacting, and delegating to one another?
@weballergy joined our podcast with @fryrsquared to explore the rise of agentic economies – and how we can diversify agent decision-making to avoid AI groupthink.
Timecodes:
00:00 Intro
1:07 Defining AI agents
4:44 Agentic exploration in science and research
15:46 Delegation between agents
22:46 Agentic security and traps
29:31 Building an agentic economy
33:22 Cognitive monoculture
36:29 Distributed intelligence· 35d@accelContext changes everything in speech recognition. Big congrats to the AssemblyAI team on this one.· 36d@MistralAIIntroducing Mistral OCR 4. It creates structure with bounding boxes, block classification, and inline confidence scores in 170 languages. 🧵👇 https://t.co/jR78NkL4xK· 36d@MistralAIMistral officially has 1,000 team members around the world!
Thank you to our incredible talent for believing in the mission and contributing to what we are building each and every day.
You too can be a part of the next chapter in our journey as we continue to grow and serve our customers. Apply for open roles at Mistral across all functions here: https://t.co/tFpS6Fcjg9· 37d@sequoia12 years ago a top scientist said AI would wipe out radiology.
Computer vision spread through every scan – and radiology demand went UP.
Hospitals hired more radiologists.
@Nvidia’s Jensen Huang on why we keep confusing a job's tasks with its purpose. https://t.co/NWhGQcAHQi· 41d@traestephensHe's baaaaaack https://t.co/do65FWakEi· 41d@astrangeThe story of how @SantiaSua is building Addi is a masterclass in scaling an AI first company. 1 in 4 Colombians now uses Addi to shop, pay, and soon to bank. Addi has 200+ agents in production - they counterintuitively started in legal! - for full automated legal, onboarding & CS.
For anyone thinking about hiring & scaling a global workforce & building agent, checkout the full conversation with me & @GEVS94 @Addi_Col· 41d@julesyooShared the AHIP 2026 stage last week with @aboehler and @peter_orszag to debate healthcare affordability + policy. Some of our takes:
🔥 Healthcare affordability has not improved in the last few years, and has even gotten worse in several areas.
🔥 We've also all aged ~20 years in the last 3 due to the rate of policy and regulatory change in gov't health insurance markets - e.g. MA v28 + new rate notice, Medicaid work requirements under HR-1, ACA subsidy cliff, etc... the health plans are fatigued (and still have a ton of work to do to actually implement and execute against these changes).
🔥 And at the same time - innovation is accelerating! From GLP-1s, cell and gene therapies, to AI-enabled care models. While these technological advancements promise profound health impact, broad affordability for them has not yet been solved. Good policy needs to address market failures related to risk pools, time horizons for underwriting said risk, and pragmatic payment models with teeth behind them.
🔥 Consumer-directed cash-pay is still a rounding error of total spend today... but we should keep an eye on the trend: consumers are waking up to the fact that they're burning $1,000s inside their own deductible for opaque, overpriced services + a miserable UX. Cash-pay-optimized suppliers are rushing into that void - and people are voting with those deductible dollars. e.g. @function, @CounselHealth, @truemed, et al are winning a rapidly growing following based on superior value and experience that the insurance-optimized system is not incentivized to deliver. Will these kinds of players eventually constitute the future version of health insurance?
🔥 In 2025, providers were out front on AI while health plans were playing defense... but in 2026, the plans are clearly leaning in, across both internal ops and external member- and provider- facing experience.
In 5 years, if we're saying healthcare affordability got better, it'll be because we finally embraced AI to bring down the cost structure of care on BOTH ends of the acuity curve. The low end is obvious. The high end - the sickest, most comorbid patients - is where we were most hesitant to say exactly what the near-term impact could be.... but therein is the grand challenge for builders: bend the cost curve in areas where the system looks the most AI-proof.· 42d@sequoiaDon't mistake task for purpose.
@Nvidia's Jensen Huang in conversation with @Konstantine: "Typing is not the job of a software engineer. Coding is not their job. Solving problems is their job."
Jensen says AI is creating more jobs, not destroying them. Why? AI lets us solve problems better than ever.· 42d@julesyoomessi - still got it. 🎩🪄· 43d@traestephenshttps://t.co/RoirVUKP3z· 44d@traestephensWow. Exciting day of soccer. https://t.co/1NEZTKJXfT· 44d@FoundersfundWill the Mafia win? Final game of the night on Thursday, 6/18. https://t.co/3nrsOh5juC· 44d@Foundersfundhttps://t.co/FpvYCd7jRV· 47d@fintechjunkieJust One More Hour
It's 11pm and you've already put in a full day. You know you should shut the laptop but there's a feature you've been thinking about since dinner. And you know that if you work with Claude Code one more hour it will exist before you go to bed.
So you stay.
Gamblers say "one more hand." Bingers say "one more episode." And now Builders everywhere are saying "Just One More Hour." A new addiction is sweeping through the Startup ecosystem.
THE MACHINE CHANGED
For all of human history, meaningful work took time. Projects unfolded over weeks, months and years. Progress was measured in milestones that were spaced far enough apart that shutting down for the night was the obviously rational choice. Nothing important was going to happen between 10pm and 8am, so why stay at the desk? Sleep was important and the work would still be there in the morning.
That math just broke.
One extra hour with Claude Code or Codex and your project is materially farther along. A few extra hours and your to-do list visibly shrinks. Features materialize and bugs get eradicated. Testing happens so quickly that you can build loops that ship, test, fix and ship again while you watch. Websites and mobile apps appear almost out of thin air. For the first time in history, a single hour of work produces SIGNIFICANT output you can see, touch and demo.
And the human brain is wired to respond to this kind of stimulus. Visible progress and accomplishment trigger dopamine. The gap between effort and reward has collapsed from months to minutes, and every Builder's reward circuitry is lighting up like a slot machine that actually pays out.
Which leads us to a question that needs to be answered: Is this dopamine earned or unearned?
EARNED VS. UNEARNED DOPAMINE
The release of dopamine creates a reward circuit in the brain that equates specific experiences with pleasure and reinforces them via intense cravings.
Cheap sources of unearned dopamine train your system to want more cheap sources of dopamine. Post on social media to chase “likes” and you get a dose of unearned dopamine. Become an “angry keyboard warrior” and the crowd’s retaliation feeds you unearned dopamine. Day trade stocks and watch the minute-to-minute fluctuation in price and you’re swimming in unearned dopamine.
Sources of unearned dopamine are everywhere. Your brain gets paid and your life gets nothing.
Earned dopamine is different. It's the reward for actual accomplishment: The deal you’ve been working on for months that closes, the marathon that you finish after a season of training, or the product that your team has been working on for months finally ships. Nobody writes thought pieces warning you about the dangers of actually finishing things. Earned dopamine is supposed to be good!
But here's an uncomfortable new truth:
THE DOPAMINE FROM JUST ONE MORE HOUR IS EARNED!
The progress is real, the feature you shipped actually works and the bug you’ve been chasing down is actually dead. This isn't a casino trick where they trick you into thinking you’re winning. You ARE winning.
And that's exactly what makes this a dangerous addiction the Startup community is facing. Nobody intervenes on an addiction that produces results. Nobody’s going to tell you to stop being productive.
COMPRESSION IS WHAT TURNS A SUBSTANCE INTO A DRUG
I try to come up with analogues to explain new conceptual frameworks, and what follows is the best I’ve come up with. It’s not great, but it’s useful in making a profound point.
The analogue: People chewed coca leaves for thousands of years without a societal crisis and the molecule was never the problem. We learned to concentrate it and speed up its delivery and the harmless substance suddenly became cocaine. Concentration plus delivery speed can turn a benign substance into a drug.
Builders have been getting high on shipping forever. But the shipping high used to come a few times a year after months of grinding, with long stretches of friction in between. AI coding tools took the same substance and removed the friction and sped everything up. The high is now available hourly, on demand, with no natural cooldown.
What nobody wants to internalize is that the friction was protective. The sheer act of typing to code, compile times and waiting on your teammates weren't useless inefficiencies. They acted as natural guardrails and forced rest into the system whether you wanted it or not. AI just removed all of them at once and we’ve embraced it without any understanding of the consequences.
THE GAME ON THE FIELD
If this were purely a personal willpower problem, the advice would be easy: Log off, touch grass and get some sleep. But the Startup ecosystem has a nasty structural twist that turns a private temptation into a collective trap.
Speed has become the new moat.
When every team has access to the same models, the same tools and the same compute, it’s tempting to believe that the only differentiator becomes how many hours you feed the machine. Your competitor's product is materially better every single morning which means shutting down at 9pm is no longer a lifestyle decision. It's a competitive decision.
I've written before about how people aren't irrational, they're rational actors responding to the game that's on the field. So take a look at the game that's now on the field for Founders: The cost of an extra hour is low, the output of an extra hour is visible, and everyone else is putting in the hour. Run that math and "Just One More Hour" isn't an addiction at all. It's the dominant strategy.
That's how arms races work. Every individual decision is rational and the collective outcome is brutal.
THE TRADE YOU'RE REALLY MAKING
So is the conclusion "work less"? No. The conclusion is that many Builders could be mispricing the trade.
When speed becomes abundant, it stops being the edge. The machine supplies the speed now. What the machine cannot supply is the judgment that turns it into velocity: Knowing what to build, what to kill, what actually matters to a customer, and what's a shiny object wearing a customer costume. Speed can drive you 100 miles an hour in circles. Speed PLUS Judgment gets you the Velocity to push your business forward. An opinionated perspective about the solution is what’s scarce in an AI-saturated Startup ecosystem, and it’s precisely this clarity of thinking that degrades for most people after midnight.
Anyone who has reviewed their own 1am decisions in the morning knows this.
A well-rested team pointing the machine at the right problems will beat an exhausted team pointing it at the wrong one, and the gap will compound every week because the machine multiplies the quality of its instructions.
THE MACHINE SUPPLIES THE SPEED. YOU SUPPLY THE JUDGMENT THAT TURNS INTO VELOCITY.
Act accordingly.
ONE MORE THING
I want to be honest about where this essay comes from. I'm back in the Builder's seat, and I live in Claude Code 10+ hours a day right now. I've always been a hard worker and I'd put my career track record of outworking the competition up against just about anyone's. I thought I understood what the pull of work felt like but this is different.
The pull of "just one more hour" is showing up for me every single night without fail, and it arrives wearing the most convincing disguise an addiction has ever worn: Genuine, demonstrable, earned progress.
I know the dopamine is real and I’m not ignorant of its long term impact. I’m very self-aware. But what I do know is that tonight, somewhere around 11pm, I'll feel the pull and work a few more hours anyway.· 48d@GreylockVCGreylock is excited to lead Niteshift’s $7M seed round, alongside @AmplifyPartners, @BoxGroup, and @svangel.
Coding agents are quickly becoming capable of taking on longer, more complex engineering tasks. But the bottleneck has shifted from what agents can write to whether they can run, test, and verify their work in real development environments.
Introducing Niteshift, the full-stack cloud for coding agents.
Niteshift lets teams run frontier and open-source coding agents inside fully configured cloud dev environments, so agents can return pull requests with evidence that they work.
@smehmood and Conor's infrastructure instincts come from years spent building at Datadog, where they worked across complex codebases, cloud development, observability, and production environments. They understand that the hard part is not just writing code, but giving software the environment, tools, and feedback loops it needs to work reliably.
We’re proud to partner with Sajid, Conor, and the Niteshift team as they build the cloud for coding agents. Read more on the Greylock blog and in TechCrunch.· 49d@julesyooI always think about this in the context of the AI-native “medical record” - and would include AI chat transcripts in there as well. Between ambient recordings + chat transcripts (effectively a form of PRO), AI care / AI care management companies will have far more health context on patients than traditional EHRs. Trad EHR notes are inherently abstractions optimized for billing versus true narratives of a) a person’s health situation and b) the clinician’s reasoning about how they decided the diagnosis and course of action.· 49d@astrangeClaude is telling me not to procrastinate https://t.co/y0PoqpWSKP· 50d@astrangeEven the best jogging strollers are hard to push and muck with your stride 🏃♀️ maybe this is the solve 😂· 52d@FoundersfundWho's lying? https://t.co/3zs09HYgeM· 54d@MistralAIWe're taking on the hardest problems in the real world 🏗️🚚 🛫⚛️
Today at The AI Now Summit, held at the Louvre, we announced AI solutions for aerospace, automotive, energy, and physics.
Deployed in production at @Airbus , @BMW, @EDFofficiel , and more.
More below: https://t.co/L6Vq7RdABg· 62d@fintechjunkieThe Score Isn't the Game
The AI personal finance race is on. OpenAI can access your bank feed with permission. It’s a great start but nowhere near enough.
I love seeing innovation, progress, and the democratization of intelligent advice in personal finance. ChatGPT and Anthropic are making waves which shouldn’t be a surprise to anyone paying attention. The category needs more thoughtful capital, more product muscle, and more shots on goal. It’s too important to ignore.
But personal finance isn’t just about budgeting or tax planning. Society has shifted in a profoundly negative way over the past decade such that financial nihilism is more the norm than the exception. People don’t feel like they can successfully navigate today’s expensive world and many people believe it’s getting worse, not better. So let me say the uncomfortable thing out loud:
Connecting your accounts isn't going to fix this.
Not even close.
And waiting for the government to fix it for you isn’t a strategy.
It’s hopium.
The Dashboard Trap
Giving credit where credit is due, the OpenAI / Hiro / Plaid integration is a real step forward. Seeing all your money in one place is genuinely better than logging into seven different apps and trying to keep a mental running total. The industry has been working on this problem since Mint launched in 2007, and a number of companies like Monarch, Origin Finance and YNAB are fantastic tools that millions of users love and use every day. ChatGPT will open up this functionality to tens of millions of new people over time which should be applauded.
But seeing your financial chaos isn't the same as fixing it. It's the equivalent of stepping on a smart scale every morning. The scale will give you useful data but it doesn't make you healthier.
Aggregation tells you the score of the game you're playing but it doesn't change the game.
The Rule Followers Are Losing
Millions of people did everything they were told to do by getting the degree, taking the stable job, saving and avoiding dumb debt. They followed the playbook their parents handed them and ticked all the boxes society told them to.
These same people are frustrated that their effort hasn’t given them the prize they thought they were chasing.
Housing prices rose 48% between 2019 and 2024. Median income rose 22%. Childcare runs 3x what HHS considers "affordable." Student loan repayment averages 20+ years and puts families behind on every major life event. The median 401(k) balance for Americans approaching retirement is a meager $65,000.
These are just facts. The numbers represent how much the social contract has changed.
I've been writing about this for a while and coined the phrase P(win) = 0 which is meant to represent the awakening people have when they believe the probability of winning is zero if they continue to follow the rules as they’ve been laid out. When they realize the math no longer maths because life became expensive and people’s incomes didn’t keep up. The conventional path stopped producing conventional outcomes a long time ago and people are frustrated that everyone keeps insisting the conventional path is the answer.
When P(win) = 0, any alternative becomes rational, even the ones that seem dumb on the surface.
It’s why 55% of Gen Z investors prefer crypto to the traditional financial ecosystem. It's why Kalshi and Polymarket went from fringe to mainstream in eighteen months. It's why 0DTE options trading has exploded. It's why "save and splurge" lives in the same wallet at the same time. People aren't being irrational. They're being rational because they feel stuck inside a game they've concluded they can't win.
Better Dashboard, Same Game
Now connect the dots with the new “connect your financial accounts” announcements.
If the underlying problem is that conventional financial advice no longer produces conventional outcomes for huge swaths of society, then aggregating bank feeds and pointing a large language model at the resulting dashboard gives you faster, prettier, more personalized versions of the same advice that already isn't working.
Calling that democratized intelligent advice is generous. It's democratized access to a playbook that's convenient but outdated. It’s great directional advice that doesn’t actually help people address root cause issues. It’s a pain killer that makes you temporarily feel better.
But the truth is harsh. A better visualization of a losing position is still a losing position.
The Real Moat
I've written before that in a world where anything can be built in a weekend, the last real moat is an opinionated perspective on the solution. I believe this is more true in the personal finance category than almost any other.
Aggregation will commoditize and so will the model layer. The edge is going to live somewhere else entirely.
It's going to live with whoever has a genuine point of view on what's actually wrong and a framework for what to do about it that respects today's realities rather than the realities of 1990.
That means saying out loud that the old playbook ("save 10%, retire at 65, 4% rule") is structurally broken for a huge portion of the country. It means building an advice layer that takes today’s reality as the starting point:
The structural gap between income and cost of living is real and can't be budgeted away
People need to live the lives they want to live now, not just plan for a hypothetical future
The goal isn't perfect optimization. The goal is making the game winnable again.
That's a product position. Most of what's being built right now is a feature position. There's a real difference.
Building It
Which brings me to what I haven't really said publicly yet.
I'm building it.
I've spent the last year quietly putting together what I think the answer looks like. At its core, it’s an AI financial workspace built on a deeply opinionated framework I've been refining that starts with today’s reality. Everything I've been writing about for years is baked into the architecture.
It's a fiduciary-minded decision engine that takes a position on what's actually broken and gives people specific, actionable moves that account for the world as it actually is, not the world we were promised.
More to come when it's ready (but we are hiring!).
I'm genuinely glad the big AI labs are wading into personal finance because the category has been underserved for decades and the work being done now will move the entire space forward. Real competition is good for users.
But the people who've been doing everything right and still aren't winning don't need another dashboard.
They need someone willing to say:
You're not crazy. The social contract was rewritten so here's how to play the game and win.
That's what I'm building. Wish me luck.
Onwards and upwards,
Fintechjunkie· 72d@natfriedmanStarted work at Meta this week. My job is to make amazing AI products that billions of people love to use.
It won't happen overnight, but a few days in, I'm feeling confident that great things are ahead.· 392d@natfriedmanCivilization is constant maintenance· 422d@natfriedmanUSA: passing 1000 state bills to slow down AI
China: passing american AI on leaderboards· 426d@Shaun_MaguireOn the "birtherday" day of @realDonaldTrump ,
Let's celebrate the best President ever - Barrack Obama... https://t.co/rNJXKlH25j· 2236d Loading terminal…