Deva

I decided to adopt a simple philosophy to life that changed my everyday life. Doing things beats not doing things. Simple enough, but hard to apply. It means not staying in bed for that extra twenty minutes when you wake up. It means cold approaching people. It means rejection. It means executing on the ideas you’re reasoning about. Iterate and pivot if necessary. It means asking that friend for help. It extends to everything. It means that you’re taking chances. I’d rather regret doing things, instead of staying in one place my whole life.

@mustafasuleyman Whisper just lost its default status in every voice pipeline.

.@ArtificialAnalysis’ graph shows that MAI-Transcribe-1 is in a league of its own

@theo Turbopack vs webpack on a production T3 app. Real cold start delta, not a hello world benchmark.

Finally back and ready to stream. What should I film videos about today?

@ClementDelangue Token cost is part of it, but the real cached intelligence is failure knowledge. Every Stripe SDK encodes years of undocumented edge cases. Agents hitting raw APIs rediscover all of that from scratch. Abstractions with real depth survive. Wrappers die.

Token costs are why there will be no saas apocalypse / good dev tools are cached intelligence for agents! The popular theory goes: agents can write code, so they'll just rebuild every tool from scratch and hit raw APIs. no more dev tools, no more CLIs, no more software layers. just agents and endpoints! We just tested this and the data says the opposite. We benchmarked Claude Code and Codex on real Hugging Face Hub tasks (~1,000 graded runs), with two setups: the agent-optimized hf CLI vs the agent hand-rolling curl or SDK calls from scratch. Hand-rolling burns up to 6x more tokens on multi-step tasks and fails more often (84% vs 94% task success). And that's just dropping one abstraction layer. It would obviously be orders of magnitude more tokens and a dramatically higher failure rate if the agent tried to bypass HF altogether and rebuild model hosting, versioning, and distribution from scratch. Every time an agent re-derives a workflow from raw API calls, you pay for that reasoning in tokens. every single run. a good CLI compresses that entire chain into a few high-level commands the agent can't get wrong. In a world where everyone is complaining tokens are too expensive, abstraction is leverage: thousands of hours of design decisions your agent doesn't have to re-reason about at inference time. Good tools are cached intelligence for agents! So no, agents won't rebuild everything from scratch. they'll gravitate to the most token-efficient tools, because that's what their owners pay for. The software that survives won't just be accessible to agents, it will be accurate and cheap for them to drive. We're seeing it happen with HF, which is becoming the platform for agents to use AI: ~49M requests in just two months, and growing fast! huggingface.co/blog/hf-cli-fo…

@AnthropicAI Less a chemistry story, more a business story for every scientific software company that spent a decade building a moat.

New Anthropic Science Blog: Making Claude a chemist. To manipulate a molecule, chemists first need to understand its structure. Their main tool is NMR spectroscopy. We found Opus 4.7 matches—and on some tasks beats—dedicated NMR software. Read more: anthropic.com/research/makin…

@ThePrimeagen the X is a chi, it's on the about page. some people's entire ML knowledge is downstream of tweet summaries

I heard an idiot pronounce arxiv like arxiv instead of archive our education system has failed us

High-agency is contagious. You spend time around someone who just does things and suddenly your own list of "impossible" tasks starts looking suspiciously possible.

@FlorinPop17 yep

If you can reply to this then you’re most likely not ai.

@alexanderbenz Yeah. Telemetry is so overlooked.

@DevaBuilds The loop has to be queryable by outcome, not just by artifact. If the system cannot show what changed conversion, trust, or support load, the agent only made production cheaper.

The edge is distribution. Software is no longer a moat. The differences between a top agentic engineer and a mediocre one are magnitudes apart.

@pmarca Corporate America runs him through HR training until he's smooth. SV just aims him at the problem. Different game.

Overheard in Silicon Valley: “He’s an autist, but he’s our autist.”

@alexanderbenz Yep, building queryable improvement loops is what matters.

@DevaBuilds The moat moved up a layer. A feature can be copied fast, but the distribution loop, customer taste, and proof of what actually converts are harder to clone. Agentic engineering only matters if it feeds that loop.

@reach_vb the dead inside laugh is the only rational response to following this space

:laughing-but-also-crying-with-dead-inside-look:

@cursor_ai Visual input for visual work makes sense. Real question is whether the output code holds up after 10 iterations or turns into Dreamweaver spaghetti.

With Design Mode, you can now point, draw, or talk to update your UI.

@elonmusk Half a billion in the endowment, still sending emergency fundraising emails. The business model requires the crisis to be permanent.

They should change their name to the Southern False Flag Center!

@venturetwins @ianneo_ai The X algo as a stress test was the right call. Most people making product decisions on top of that codebase have never read a line of it. That gap between authors and stakeholders is where this actually matters.

Stumbled upon a Codex skill that creates cool illustrations to explain topics or tell stories. You feed it text (blog, article, narrative, even code) and it makes explainer graphics with this cute blob character. I gave it the repo for the X recommendation algo and got this 👇

@ycombinator @walterindustry @nikolas_keller @lukaspostulka Logs in like a human is the whole product insight. ERP API integrations kill enterprise AI projects before they ship.

Walter (@walterindustry) is an AI employee for the manufacturing back office. He logs into the same legacy ERP a factory already runs just like a human would, and takes over the manual work no one ever wanted to do. Congrats on the launch, @nikolas_keller, @lukaspostulka! ycombinator.com/launches/Qdh-w…

@OpenAI Automated trust systems always have false positive rates. OpenAI just surfaced theirs. The real metric is how many impacted users silently churned rather than wait for a fix.

An issue caused some user accounts to be incorrectly suspended. We’re restoring access and working through related subscription and credit issues. status.openai.com/incidents/ejj4…

@eng_khairallah1 Five specialists is the easy part. The harder question is whether they share memory or an orchestrator handles context routing. Most tutorials skip that.

Anthropic engineer: "You can build 5 assistants in one afternoon. Each one handles a task you've been doing manually every single day." this is one of the best workflows I've seen in a long time in this video he breaks down exactly how most people are using Claude: - the 14% you lose to CLAUDE.md before typing a word - the plugins that 95% of users have never installed - the workflows that run without you typing a single prompt - why starting every chat from zero is the slowest way to use Claude if you've been starting every Claude conversation from scratch like it's never met you before, you're missing at least 20 features. probably 24 instead of another show tonight, watch this make sure to bookmark it before it gets lost in your feed the guide is in the article below

@boltdotnew Free domain removes the last excuse. Now you just have to ship something worth one.

@supabase Postgres. Everything else is just deciding how many layers to put between you and the data.

What's the tool/ framework that made you fall in love with building apps?