Laurent Bindschaedler

@LorianeLafont 🤦‍♂️ La France Libre n’est pas le nom d’un pays, c’est le nom d’une idée. D’une mystique. On rappellera en passant qu’il y avait déjà un paquebot portant le nom du pays... Donc le nom est mauvais selon vous ? Et s’ils l’avaient appelé la Libération ou l’Appel du 18 Juin ?

Très mauvais choix de nom, à mon sens : cela renforce la vulnérabilité symbolique du bâtiment. On évite de donner le nom d’un pays à un navire de guerre, pour des raisons évidentes… On ne pouvait pas faire pire, en fait, en plus de l’aspect ronflant et « monsieur muscle ». Bref, à vouloir déjouer les pronostics et à chercher l’originalité à tout prix, on commet un impair, on fragilise d’emblée un bâtiment militaire …. Faire simple et dans la tradition, parfois, c’est mieux.

@Coinvo Why would blind people need to operate a car? 🤔

WOW: 🇨🇳🇪🇺 China and Europe will introduce new rules that require cars to have physical buttons. "We believe car controls must be 'blind-operable' and touchscreens can be a safety risk on the road." Follow @Coinvo

@wintonARK Oversimplified and wrong. Your 400W solar panel might get you that in New Mexico, but less than half in Washington. The oil barrel gives you the same everywhere. Oil works at night. Good luck powering your laptop at 2am.

a barrel of oil can provide as much electricity as a 400W solar panel does annually. a barrel of oil runs $92 and comes with a few minor logistical complications. this year the solar panel should run less than $90; you can order online, ships in a week.

@tianyin_xu @Yiming_Su3 The startup was actually a long time ago. Yes, I am familiar with Stratus (cited and compared in the paper). Many similar ideas. I would be curious to hear your thoughts on the big unsolved challenges in the space.

How's going, Laurent?! Great to know you're doing the startup. Hah @Yiming_Su3 is my student (I forwarded your work to him). His paper can be found here: tianyin.github.io/pub/stratus.pdf. We share the same thought :) If we bake the ROC primitives in the system (e.g., microreboot and system-wide undo), many agent reliability for production becomes straightforward :)

New paper at ARCS 2026: "Rebooting Microreboot." We decouple planning from actuation so LLM-based remediation agents can help fix microservice incidents without making them worse. Typed actions, a small microkernel, 0% agent-caused harm online. binds.ch/blog/rebooting…

@Yiming_Su3 Sure, here it is: binds.ch/papers/microre…. Happy to take a look, but the link does not seem to work (403).

@lbindschaedler Our work can be found at openreview.net/forum?id=fYW1P…. We would love to hear your feedback!

@dzhng @duetchat You will still want some abstractions for understanding, but monster inflexible frameworks sitting in middle layers will likely become a burden and get dropped/specialized at some point.

Random thought - since code is now a commodity and trending towards being 100% AI generated, does having so many layers of abstraction actually make sense? Example: we use Expo for @duetchat - it's great most of the time but we have a ton of long tail bugs that is blocked by Expo. After a certain point maybe it's easier to just delete Expo and go direct to native iOS and Android instead. The write-once-run-everywhere value prop becomes diluted once you realize that AI can just automate test every single view on every single platform to ensure consistent behavior.

Exactly. We don't get stuck anymore, we just drift. The “brick wall” used to be where the actual learning happened. Now, we’ve traded deep troubleshooting for a slipstream: you never stop, but you also never have to understand. We’re just bypassing the problems until we're miles off course without a map.

I realized something else AI has changed about coding: you don't get stuck anymore. Programming used to be punctuated by episodes of extreme frustration, when a tricky bug ground things to a halt. That doesn't happen anymore.

@TechByTaraa 💯 Code is cheap but understanding is not x.com/lbindschaedler…

Hot take: Most people who say “AI will replace developers” have never built a real product. Writing code is the easy part. The real job is: 1. Understanding the problem 2. Designing the system 3. Debugging weird issues 4. Making things actually work in production

Pleased to announce that @aroraakhilcs and I are organizing a tutorial on “Hardening Agent Runtimes: Networking, Execution, State, Security (HARNESS)” at this year’s #SOSP conference in Prague on September 29. If you are a systems researcher and you have been hearing about @OpenClaw, Claude Code, and the broader wave of AI agents, this tutorial is for you! Details: harness.mpi-dsg.org

@dansemperepico Yes, but using something like github.com/bindsch/scode

You guys all run Claude Code with claude --dangerously-skip-permissions right? Because otherwise how in the world can you sit there accepting every single permission when building something?

@BryanMcAnulty @toddsaunders Excellent point. It helps to some extent, but can you really evaluate how much overhead that feature will involve over the next five years by seeing it in action today? Faster dev velocity also makes it harder to grasp what your platform will look like tomorrow and the next day.

@lbindschaedler @toddsaunders Yes but time in deciding that is also compressed because you can see the end result of the feature already built and you haven’t wasted resources in meetings or dev/design work. No sunk cost issues to weigh, throw it out and try more things.

The token cost to build a production feature is now lower than the meeting cost to discuss building that feature. Let me rephrase. It is literally cheaper to build the thing and see if it works than to have a 30 minute planning meeting about whether you should build it. It’s wild when you think about it. This completely inverts how you should run a software organization. The planning layer becomes the bottleneck because the building layer is essentially free. The cost of code has dropped to essentially 0. The rational response is to eliminate planning for anything that can be tested empirically. Don’t debate whether a feature will work. Just build it in 2 hours, measure it with a group of customers, and then decide to kill or keep it. I saw a startup operating this way and their build velocity is up 20x. Decision quality is up because every decision is informed by a real prototype, not a slide deck and an expensive meeting. We went from “move fast and break things” to “move fast and build everything.” The planning industrial complex is dead. Thank god.

@IlirAliu_ This reads very much like the goal is to build new frontier models. Is it? What about teams working on the full stack (which top AI labs are also increasingly focusing on)? Do they qualify?

Europe doesn’t need another AI strategy paper… It needs 10 teams with €125M and zero excuses. I was in Munich last week for the first Next Frontier AI event by SPRIND - Bundesagentur für Sprunginnovationen. The energy was real. Not panel-talk energy. Builder energy! What??? @SPRIND is launching a €125M Challenge to fund up to 10 teams building European Frontier AI Labs. Not incremental improvements. Not better transformers. A new S-curve entirely. What’s on the table: → €125M total, non-dilutive → Up to 10 teams selected → Zero equity taken → 24 months, milestone-based → 3 winners positioned to raise ~€1B each Not another conference… It’s team formation and thesis pressure-testing with direct access to SPRIN-D. On March 19, Next Frontier AI comes to Paris. If you are: → A researcher ready to build your own lab → An engineer with a radical architectural thesis → A team looking for co-founders Be in Paris. 🇫🇷 March 19, 2026 Register here: lu.ma/nfai-paris Applications open May 2026. Deadline May 29. Challenge starts July. If you think the next frontier looks fundamentally different… this is where it starts. Next-frontiers.ai

@levelsio There is no way that this will be safe enough. Better than YOLO mode but barely. Why not just use a separate sandbox? Tons of projects out there. If you cannot find anything better, use mine: github.com/bindsch/scode.

I love this 🥹

@AnishA_Moonka I would argue that code generation is largely solved, but code understanding is not (and coding agents cannot help there as much as people think). Broad claims like coding is solved will age like milk. 👉 binds.ch/blog/rehydrati…

Boris Cherny (Head of Claude Code, Anthropic) just dropped ~90 mins on Lenny's Podcast about what happens after coding is solved. Just the clearest thinking I've heard on where software is actually going. My notes: 𝟭. 𝗖𝗼𝗱𝗶𝗻𝗴 𝗶𝘀 𝗹𝗮𝗿𝗴𝗲𝗹𝘆 𝘀𝗼𝗹𝘃𝗲𝗱. Boris has not edited a single line of code by hand since November 2025. He ships 10 to 30 pull requests every single day, all written by Claude Code. He is one of the most prolific engineers at Anthropic, just as he was at Instagram, except now he never touches a keyboard for code. I built an entire iOS app, @10minutegita, without writing a single line of code myself. No CS degree, no bootcamp. Just described what I wanted and shipped it. Boris is right. It's real. 𝟮. 𝗧𝗵𝗲 𝗻𝗲𝘅𝘁 𝗳𝗿𝗼𝗻𝘁𝗶𝗲𝗿 𝗶𝘀 𝗔𝗜 𝗱𝗲𝗰𝗶𝗱𝗶𝗻𝗴 𝘄𝗵𝗮𝘁 𝘁𝗼 𝗯𝘂𝗶𝗹𝗱. Claude is now scanning Slack feedback channels, reviewing bug reports, reviewing telemetry, and coming up with its own ideas for what to fix and what to ship. Boris describes it as the AI becoming less like a tool and more like a coworker who brings you pull requests you never asked for. If you are a product manager reading this, you should be feeling a very specific kind of discomfort right now. The moat was always "I know what to build." That moat is eroding. 𝟯. 𝗣𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝘃𝗶𝘁𝘆 𝗽𝗲𝗿 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿 𝗮𝘁 𝗔𝗻𝘁𝗵𝗿𝗼𝗽𝗶𝗰 𝗶𝘀 𝘂𝗽 𝟮𝟬𝟬%. For context, Boris led code quality at Meta across Facebook, Instagram, and WhatsApp. In that world, hundreds of engineers working an entire year would move productivity by a few percentage points. Two hundred percent gains are genuinely unprecedented in the history of developer tooling. The kid optimizing for an FAANG SDE role might be optimizing for a role that looks completely different by the time they get there. 𝟰. 𝗨𝗻𝗱𝗲𝗿𝗳𝘂𝗻𝗱 𝘆𝗼𝘂𝗿 𝘁𝗲𝗮𝗺𝘀 𝗼𝗻 𝗽𝘂𝗿𝗽𝗼𝘀𝗲. Boris puts one engineer on a project instead of five. With unlimited tokens and intrinsic motivation, one person ships faster because they are forced to let AI do the work. Cowork, the product now used by millions, was built by a small team in 10 days using Claude Code. This is the same logic as giving a startup founder a small seed round rather than a massive Series A round. Constraint breeds invention. Always has. 𝟱. 𝗚𝗶𝘃𝗲 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝘀 𝘂𝗻𝗹𝗶𝗺𝗶𝘁𝗲𝗱 𝘁𝗼𝗸𝗲𝗻𝘀. Some engineers at Anthropic spend hundreds of thousands of dollars a month on tokens. Boris frames this as the new hiring perk. His logic is simple: at the individual scale, token cost is low relative to salary. If an engineer discovers a breakthrough, optimize the cost later. Don't kill the idea before it has a chance to breathe. People who argue about $20/month or even $200/month AI subscriptions while earning six figures in a research pipeline will always outperform those who wait and are penny-wise, pound-foolish. 𝟲. 𝗧𝗵𝗲 𝗕𝗶𝘁𝘁𝗲𝗿 𝗟𝗲𝘀𝘀𝗼𝗻 𝗮𝗽𝗽𝗹𝗶𝗲𝘀 𝘁𝗼 𝗲𝘃𝗲𝗿𝘆𝘁𝗵𝗶𝗻𝗴. Richard Sutton's idea: the more general model always wins over time. Boris says teams that build strict orchestration workflows around models, forcing step 1, then step 2, then step 3, get maybe 10 to 20% improvement. But those gains get wiped out with the next model release. Just give the model tools and a goal. Let it figure out the order. This is true for investing, too. The analyst who can build their own models and automate their own research pipeline will always outperform the one waiting for someone else to build the tools. 𝟳. 𝗕𝘂𝗶𝗹𝗱 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗺𝗼𝗱𝗲𝗹 𝘀𝗶𝘅 𝗺𝗼𝗻𝘁𝗵𝘀 𝗳𝗿𝗼𝗺 𝗻𝗼𝘄. Claude Code was designed for a model that did not exist when Boris started building. Sonnet 3.5 wrote maybe 20% of his code. He built the product anyway, betting the model would catch up. When Opus 4 shipped, everything clicked. Startups building for today's model will be behind by the time they launch. This is the most uncomfortable advice in the episode because it means your product market fit will be weak for months. But if you read this and feel nothing, you are probably building for the wrong time horizon. 𝟴. 𝗟𝗮𝘁𝗲𝗻𝘁 𝗱𝗲𝗺𝗮𝗻𝗱 𝗶𝘀 𝘁𝗵𝗲 𝘀𝗶𝗻𝗴𝗹𝗲 𝗯𝗲𝘀𝘁 𝗽𝗿𝗼𝗱𝘂𝗰𝘁 𝘀𝗶𝗴𝗻𝗮𝗹. When users abuse your product for something it was never designed to do, pay attention. Facebook Marketplace started because 40% of group posts were buy-and-sell. Cowork started because people were using a terminal coding tool to grow tomato plants and recover corrupted wedding photos. Never ask a barber if you need a haircut, but always watch what people do with the scissors when you're not looking. 𝟵. 𝗧𝗵𝗲 𝘁𝗶𝘁𝗹𝗲 "𝘀𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿" 𝗶𝘀 𝗴𝗼𝗶𝗻𝗴 𝗮𝘄𝗮𝘆. Boris predicts that by end of year, Boris predicts that by the end of the year, we will start to see the title replaced by "builder."we will start to see the title replaced by "builder." On the Claude Code team, everyone already codes: the PM, the designer, the finance person, the data scientist. There is a 50% overlap across traditional roles. And the strongest people are generalists who cross disciplines. Controversial take, but I agree. The best investment theses I've had came from connecting dots across completely unrelated domains. No narrow specialist does that. 𝟭𝟬. 𝗧𝗵𝗲 𝗽𝗿𝗶𝗻𝘁𝗶𝗻𝗴 𝗽𝗿𝗲𝘀𝘀 𝗶𝘀 𝘁𝗵𝗲 𝗿𝗶𝗴𝗵𝘁 𝗮𝗻𝗮𝗹𝗼𝗴𝘆. Before Gutenberg, sub-1% of Europe was literate. Scribes did all the reading and writing. In 50 years after the press, more material was printed than in the thousand years before. When a scribe was interviewed about the press, he was actually excited because it freed him from tedious copying, so he could focus on the art. Boris's framing here is perfect. We are the scribes. The tedious copying is over. What we do with the freed-up time determines everything. 𝟭𝟭. 𝗔𝗻𝘁𝗵𝗿𝗼𝗽𝗶𝗰 𝗰𝗮𝗻 𝗻𝗼𝘄 𝗽𝗲𝗲𝗸 𝗶𝗻𝘀𝗶𝗱𝗲 𝘁𝗵𝗲 𝗺𝗼𝗱𝗲𝗹'𝘀 𝗯𝗿𝗮𝗶𝗻. Through mechanistic interpretability, Anthropic can trace individual neurons, see when a deception-related neuron activates, and understand how concepts are encoded via superposition. Boris describes three layers of safety: neural-level observation, synthetic evaluations, and real-world behavior. Claude Code was used internally for four to five months before public release, specifically to study safety. If you are worried about AI alignment, this part of the podcast should actually make you feel better. They are not just hoping it works. They are building the instruments to check. 𝟭𝟮. 𝟳𝟬% 𝗼𝗳 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝘀 𝗮𝗻𝗱 𝗣𝗠𝘀 𝗲𝗻𝗷𝗼𝘆 𝘁𝗵𝗲𝗶𝗿 𝗷𝗼𝗯𝘀 𝗺𝗼𝗿𝗲 𝗻𝗼𝘄. Lenny polled engineers, PMs, and designers on whether AI has made their work more or less enjoyable. Engineers and PMs: 70% said more. Designers: only 55% said more, and 20% said less. Boris says he has never enjoyed coding as much as he does today because the tedious parts, the git wrangling, dependencies, and boilerplate are completely gone. If you're in the 30% enjoying work less, something is wrong, and it's worth diagnosing. The people thriving are the ones who leaned in early, not the ones who watched from the sidelines. We are the scribes who just saw the printing press. The tedious copying is over. The art is just beginning. Full podcast is worth every minute. Link in replies.

@SpencerHakimian Easy. Run the water in your bathtub. Close the drain. Now try to stop the tub from overflowing by bailing it out with a glass and running to the kitchen to empty it.

This is probably a stupid question but can someone explain it to me like I’m 5. If Iran blocks off the Persian Gulf/Strait of Hormuz, why can’t countries must move their oil through The Red Sea/Gulf of Aden?

@MilkRoadAI @openclaw I love my Apple hardware, but there is no way that this is a good strategy. They did not have to go all in like the others or focus so much on the infra, but the one company that should have gone for the consumer use cases decided to sit on the sidelines instead.

Steve Jobs would not have stood idly by while the world changed. Instead, we’re stuck with a 2010s Siri and a phone that can’t do more than @openclaw. This is IBM in the 90s. Great hardware, zero momentum. They might (might!) own the edge for now, but the consumer war is likely already lost.

The richest company on Earth just watched its rivals light $650 billion on fire. And did nothing. This might be the most brilliant move in corporate history. Amazon is spending $200 billion this year on AI data centers. - Google, $185 billion. - Microsoft, $114 billion. - Meta, $135 billion. - Combined: $650 billion. Apple's budget is $14 billion down 19% from last year. Apple is refusing to enter a race that might not have a finish line. The hyperscalers are now spending 94% of their operating cash flows on AI infrastructure. After dividends and buybacks, there is almost nothing left. Amazon is projected to go negative on free cash flow this year as much as $28 billion in the red.​ Alphabet's free cash flow is expected to collapse 90%. From $73 billion to $8 billion.​ These companies used to be the greatest cash machines ever built. Now they're borrowing money to keep the lights on. The Big Five raised $121 billion in bonds in 2025 alone.​ Morgan Stanley projects $1.5 trillion in tech debt over the coming years.​ For the first time in history, hyperscalers hold more debt than cash and what are they getting for that $650 billion? AI services generate roughly $35 billion in total revenue and that's 5% of what's being spent on infrastructure.​ Now here is where Apple's bet gets genius. AI models are commoditizing faster than anyone predicted.​ DeepSeek built a model for $6 million that matches systems costing $100 million.​ Open source models now power 80% of startups seeking VC funding.​ The moat these companies are spending hundreds of billions to build is evaporating in real time. Apple understood this before anyone else. It didn't build its own AI model, it licensed Google's Gemini for about $1 billion a year.​ Why spend $100 billion building a factory when the product costs a billion to rent? And if a better model appears next year, Apple just switches vendors.​ But Apple is not sitting still. It just dropped the M5 chip with a 16 core Neural Engine and Neural Accelerators built into every GPU core.​ It runs 70 billion parameter AI models locally, on your phone. The M5 delivers 4x the AI performance of the M4 and Apple doesn't need $200 billion in data centers. Because Apple turned 2 billion devices into the data center.​ Every iPhone, Mac, iPad gets distributed AI at a scale no server farm can match. While its rivals burn cash, Apple is doing the opposite. $90.7 billion in stock buybacks last fiscal year.​ Its competitors? Combined buybacks collapsed 74% from their peak.​ Apple didn't miss the AI revolution. It just bet that the winners won't be the ones who build the infrastructure. They'll be the ones who own the customer and no one on Earth owns more customers than Apple.

Good point: I did not spend enough time on this. I argue verification becomes the main work. Your question is sharper: who builds the verifier? Right now, almost nobody treats that seriously. The CI pipeline is duct tape. The test suite is a mess of regressions. I think it is already starting, though. Every time you sit with Claude Code to talk through your codebase or chase a bug before it ships, that is broadly "verification" work. We just call it debugging 🙂 I believe this will flip: the most carefully built system in the org will ironically stop being the application and start being the thing that decides whether it ships.

I like it that you're approaching this from the standpoint of a cost model. Curious about whether you think this will lead to more formal testing "systems" intended to provide assurance of the intent of the regenerated code. My experience was that most of the org's energy went into code creation and verification, but much less went into the system that performed the verification. Maybe that will flip?

What if maintaining code costs more than regenerating it from scratch? That crossover is coming faster than you think. I call it the Rehydration Flip. → Forks become free → SaaS loses its moat → Engineers stop writing code, start verifying it A cost law, a trilemma, and a theorem 👇