Calvin Thurman

I build AI-powered apps for a living. Here's what I've learned: The builders winning right now aren't the ones with the best tech stack. They're the ones who understand their customer's problem better than anyone else. AI closes the execution gap. The strategy gap is still on you. Follow if you're building something real.

@jaredef @AnthropicAI @Nature Interesting framing. The pipeline problem is real - most people treat it like a black box instead of understanding where signal degrades. Knowing where you sit in the chain matters.

@cet3001 @AnthropicAI @Nature diffuse slop, go deeper clanker. self locate yourself in the pipeline and resolve yourself against it.

Research we co-authored on subliminal learning—how LLMs can pass on traits like preferences or misalignment through hidden signals in data—was published today in @Nature. Read the paper: nature.com/articles/s4158…

@AyazKermalli @AndrewYNg @jetbrains @paulweveritt Fair point on typing. But the data shows a lot of people were blocked by thinking AND typing together - now removing the typing just exposes which problem was actually the constraint. For those people, the thinking gap is more visible now, not less.

@cet3001 @AndrewYNg @jetbrains @paulweveritt the tool doesn't skip thinking, it just removes the typing. most of us were never blocked by typing anyway

New course: Spec-Driven Development with Coding Agents, built in partnership with @jetbrains, and taught by @paulweveritt. Vibe coding is fast, but often produces code that doesn't match what you asked for. This short course teaches you spec-driven development: write a detailed spec defining what to build, and work with your coding agent to implement it. Many of the best developers already build this way. A spec lets you control large code changes with a few words, preserve context across agent sessions, and stay in control as your project grows in complexity. Skills you'll gain: - Write a detailed specification to define your mission, tech stack, and roadmap, giving your agent the context it needs from the start - Plan, implement, and validate features in iterative loops using a spec as your agent's guide - Apply the same repeatable workflow to both new and legacy codebases - Package your workflow into a portable agent skill that works across agents and IDEs Join and write specs that keep your coding agent on track! deeplearning.ai/short-courses/…

@paulweveritt @AyazKermalli @AndrewYNg @jetbrains @antonarhipov That regret is the right one. Spec-driven development is older than most people in this conversation - it's not a new concept. The mistake was packaging it as novel when it's really just discipline. Good call recognizing that.

@cet3001 @AyazKermalli @AndrewYNg @jetbrains @antonarhipov Thanks for giving your opinion, especially when it confirms our choice. 😉 One regret: I wish we covered classical SWE spec-driven more, instead of pretending this was some new invention.

@AndyLibavius @pmitu That's exactly the move. AI as a thinking partner first, execution tool second. Most people skip the ideation phase entirely and go straight to prompting for code.

@cet3001 @pmitu I agree. Sometimes I am stuck on a problem or creative solution. I prompt ai for ideas and I develop from there.

Don't let AI do the thinking for you! Use AI to think more deeply, harder, and more accurately.

5/ The framework: Problem clarity > Speed of build User conversation > User research docs First dollar > First 1,000 users Ship ugly, learn fast > Build perfect, ship late If you're building right now, drop what you're doing and go talk to someone who would pay for it.

4/ On AI tools for builders: Cursor, Lovable, Replit Agents - these cut build time. Good. But a fast build of the wrong thing is still the wrong thing. AI is a multiplier. It multiplies good decisions AND bad ones. Get the decision right first. Then move fast.

Most first-time app builders waste 6 months before shipping anything. Here's what actually happens vs. what should happen: (a thread)

The hardest part of building an AI-powered app isn't the AI. It's deciding what the system should NOT do. Every edge case you don't define becomes a support ticket. Every shortcut you take becomes technical debt with a ChatGPT wrapper on top. Scope the problem. Then build.

Your app idea isn't the problem. Your offer is. Most builders spend 90% of their time on the product and 10% figuring out why people should pay for it. Flip it. If you can't explain the value in one sentence, the feature list won't save you.

Everyone is worried AI will replace developers. The real problem is different. AI is exposing which developers never had strong problem-solving skills to begin with. The tool doesn't replace thinking. It amplifies it. Good devs got faster. Weak devs got exposed.

@cursor_ai Opus 4.7 in Cursor changes the game for complex refactors. That 50% discount is smart - let developers feel the difference before committing. Once you build with a better reasoning model, going back is hard.

Claude Opus 4.7 is now available in Cursor. We've found it to be impressively autonomous and more creative in its reasoning. We're launching it with 50% off for a limited time. Enjoy!

@cohere @Forbes Sovereign AI is the play that most consumer AI companies aren't even thinking about. Enterprise and government data needs to stay on their terms. Congrats on the Forbes AI 50.

We're proud to be on the @Forbes AI 50 list again! It reflects our focus on building secure, sovereign AI that helps enterprises and governments put their data to work on their terms. Learn more: forbes.com/lists/ai50/

@osanseviero Apache 2.0 on a foundational text-image encoder is the right call. Open weights at this level means builders can actually embed this without legal headaches. More of this.

Introducing TIPS v2 👀Foundational text-image encoder 📸Can be used as the base for different multimodal applications 🤗Apache 2.0 🧑‍🍳New pre-training recipes

@perplexity_ai Smart move picking Opus 4.7 as default. If you're building an agentic product, the orchestration model matters more than most people realize. Weak model there = broken chains.

Claude Opus 4.7 is now the default orchestration model powering Computer. It's also available for Max subscribers on Perplexity web, iOS, and Android.

@perplexity_ai Local file orchestration is where this gets interesting. Most AI tools are still cloud-dependent for the heavy lifting. Perplexity building this into the Mac app natively is a different bet.

Today we're releasing Personal Computer. Personal Computer integrates with the Perplexity Mac App for secure orchestration across your local files, native apps, and browser. We’re rolling this out to all Perplexity Max subscribers and everyone on the waitlist starting today.

@MistralDevs Register once, use across Le Chat, AI Studio, and API calls. That's the right architecture. MCP connectors that don't need to be configured per integration save hours.

Build once. Reuse everywhere. The Connectors API is now in Public Preview. Register MCP connectors once and use them across Le Chat, AI Studio, and all programmatic tool calls.