Aaron Levie diagnoses a specific executive pathology: CEOs are uniquely prone to AI psychosis because they are far enough from the last mile of work that they only see the happy path. They prototype a product but never review the code for production. They generate a contract but never verify the terms or wire up legacy systems.

His prescription is blunt: the best thing a CEO can do is use AI a ton personally, to come out the other side with an appreciation for both the upside and the real work that goes into making agents sustainable in the enterprise.

With 4865 likes, this clearly struck a nerve across the entire tech industry.

This thread complements Levie's diagnosis with the organizational mechanics: CEOs feel AI FOMO but lack the muscle for hands-on work, so they set sweeping imprecise mandates. Employees respond with performative low-effort demos to meet coarse goals.

The predictable outcome: two years pass with no real progress, and a startup with hands-on leadership disrupts you.

Peter Yang shares a quote from a founder who raised $2M but isn't hiring: as a founder you want to understand every job before you hire for it, and agents are faster to onboard and retain all training so you can improve them over time.

This is not «AI replacing workers» framing. It is «use agents to deeply understand a role before hiring a human for it»—a genuinely new operational pattern.

Zara Zhang shares a vignette that captures the IC-to-manager pendulum swinging back: an engineering manager she ran into just voluntarily transitioned back to being an IC and has never been happier, because she can finally get her hands on building again.

Combined with her observation about OpenAI open-sourcing Codex, the picture is clear: the people closest to the tools are the ones having the most fun right now.

TRQ's takeaway from the Bun rewrite is counterintuitive: legacy codebases are about to become incredibly valuable—not as running software, but as a source for distilling code into new forms.

His logic: every game should be crossplatform, all legacy software should work on the web, and we don't need COBOL anymore. If models can ingest a massive legacy codebase and translate it into something modern, the codebase itself becomes a training asset, not a liability.

With 1413 likes and 133 likes on the follow-up clarifying that models aren't quite there yet but will get there, this thesis clearly resonated.

Nikunj Kothari shares a detailed workflow for using Claude Code to reverse-engineer any website's API. The trick: point Claude Code at browser_harness or Playwright, ask it to sniff network requests, click around the sections you want, and let it figure out the API structure, auth requirements, and rate limits from the logs.

He uses this for side projects, monitoring, and scraping. His forward-looking claim: every website will soon need to be headless, we'll need mechanisms for agents to pay websites programmatically, and just as we have llms.txt for data, we'll soon need tools.txt for agents.

Garry Tan distills the current moment into one line: high agency high taste is the unlock these days. When the tools level the playing field for execution speed, what differentiates outcomes is the ability to act independently and the judgment to know what good looks like.

Guillermo Rauch keeps it simple: How do you build a great brand? Build a great product. In an era where AI lets anyone ship fast, the brands that endure will be the ones backed by genuinely good products, not just good marketing.

Zara Zhang notes a meaningful shift: OpenAI is becoming more open, and Codex is now open source. For the builder community, this matters because it lowers the barrier to building on top of these tools and contributing back.