AI advantage is shifting from raw model scale to deep integration with unique organizational data, workflows, and human expertise through continual learning loops. Ideally, this will result in distributed, defensible value rather than winner-take-all dynamics around general models. We're fully aligned around building the tools necessary (world-class inference, post-training, etc.) to make this a reality for every enterprise.

Claude 5 Fable (high) “Make a Minecraft clone” I’m stunned.. it made this in 20 minutes, one shot. Multiple Biomes, day time/night time, different ores, Caves & more!

Fable is a very impressive model. But it is also - Expensive - Slow - Rate limited - Nerfed for important R&D tasks - Not private (retains prompts/responses even for enterprise users) The difference between owned and rented intelligence gets clearer by the day.

Anthropic is calling for top AI labs to weigh slowing the pace of development, suggesting that AI systems are advancing so rapidly that they may soon be able to improve themselves without human intervention in ways that could pose societal risks. on.wsj.com/4ulkmFh

Second for second, @tylercowen packs more substance into a talk than anyone I'm aware of. This is a clear, non-hysterical, and somewhat soothing discussion of our AI future.

it's important, i think, to resist the temptation to ship garbage

People freaking out over my AI spend. What nobody sees: Part of what excites me so much about working on OpenClaw is that I'm trying to answer the question: How would we build software in the future if tokens don't matter? We constant run ~100 codex in the cloud, reviewing every PR, every issue. If a fix on main lands, @clawsweeper will eventually find that 6 month old issue and close it with an exact reference. We run codex on every commit to review for security issues (as it's far too easy to miss). We run codex to de-duplicate issues and find clusters and send reports for the most pressing issues. We have agents that can recreate complex setups, spin up ephemeral crabbox.sh machines, log into e.g. Telegram, make a video and post before/after fix on the PR. There's codex that watch new issues and - if it fits our documented vision well, automatically create a PR of it. (that then another codex reviews) We have codex running that scans comments for spam and blocks people. We have codex instances running that verify performance benchmarks and report regressions into Discord. We have agents that listen on our meetings and proactively start work, e.g. create PRs when we discuss new features while we discuss them. We build clawpatch.ai to split all our projects into functional units to review and find bugs and regresssions. We do the same split for security with Vercel's deepsec and Codex Security to find regressions and vulnerabilities. All that automation allows us to run this project extremely lean.

Interesting article on treating agent output like compiler output (and why) skiplabs.io/blog/codegen_as_…

I worry deeply already about companies controlling access to very powerful AI, which will come in a soft form with very expensive subscriptions. This is a step further, with the government confusingly exerting control without clear explanation. This control of AI can create massive dystopian societies. It’ll rapidly lead to concentration of power. Having open models follow closely in capabilities is a great way to minimize political and power games here.

All the best programmers I know are starting to write code by hand again

Talking to smarter folks than me, I'm convinced many of the AI folks in my timeline are full of shit. Nobody is "running 20 agents over night" and building stuff for actual users. Maybe some are building internal tools or disposable software. Maybe. But building software people like using? That doesn't get hacked on day one or blow up after the 3rd user? Nope. I don't even understand what that's supposed to look like. Do you work out a 57 pages document that perfectly describes what you want to build and then summon 14 agents and have them run wild for 6 hours? And what comes out on the other end isn't a broken pile of shit? Nope. Not buying it. PS: it may also be that I have an IQ of 82 and can't figure it out.

It's crazy to me how much of software development is 'finding the right way to describe something that doesn't exist yet'. It's a language problem. Figure out the right nouns and verbs, and you can build a system. And LLM's are GREAT at helping you with this.

From now on, nobody should be allowed to say anything about AI unless they've watched @badlogicgames's talk: youtube.com/watch?v=RjfbvDXp… This slide in particular is gold. There are good ways to use coding agents and there are bad ways. These are the good ways! Other people on Twitter are gradually reaching the same conclusions, so save yourself some time and just watch Mario's talk. (I don't think this list will fundamentally change with future models, either — even if the benchmarks reach 100%. Disagree? Tell me why.)

I've been using Claude Code exclusively for 6 months and I'm still not convinced on this whole AI thing. There are some *seriously* insidious problems that worry me, and I don't see them being fixed any time soon. Every release of a new model, I see hundreds of posts where people think because they one-shotted X or Y, software jobs are cooked (I've probably made one or two of these posts myself). But none of those examples are actually representative of real-world software. If I set it to work on an ambiguous or highly complex problem that has a lot of branching in the solution space, I've noticed the following: - It can often generate a working solution in one-shot, which gives me a false sense of confidence that the AI knows exactly what it's doing. - As I continue to work the problem, I've noticed the AI will start to narrow its focus more and more, not considering how a fix or solution plays into the big picture. - The quality of a solution depends on *how* I prompt it, which is really, really bad. Software engineering should be deterministic, not a dice roll. - It will often ignore instructions I have explicitly stated in the rules file, which removes any confidence I have in the code it generates. - It consistently overstates its confidence in a solution. I literally just got this response from Claude: "I overstated that. Honest answer: it depends on the scene and implementation; the 2–4× figure was too confident." If I had never pushed back, I would have been operating on incorrect information. - It is far too agreeable. If I'm not careful in my wording, the AI will blindly follow my instructions, even if they are suboptimal. I want a real coding partner that challenges my ideas, not an ass-kisser. Don't get me wrong—AI has helped me build some amazing things faster than I ever could without it. But the more I use it, the more I begin to question the direction things are headed. If the AI was more direct about what it (not) capable of, it'd be a lot easier to work with. But being gaslit every step of the way makes the process stressful as hell. Going back to manual coding isn't even an option since the value of having AI *potentially* generating the correct code in 1/10 or 1/100 of the time is literally too good to pass up on. Sorry for the rant, drank way too much cold brew this morning.