Peter Yang passes along a lesson from Ryan Carson that inverts the old startup playbook: what used to feel like procrastination—building systems instead of shipping the MVP—is now a prerequisite for shipping effectively with AI agents.

Carson's quote: 'We used to say just do the bare minimum to get the MVP out. Don't spend time on systems. It's literally reversed now. You have to spend a lot of time setting up your documentation. Build all that into a cron job with a skill file, and suddenly you're doing the work of 10 people.'

This is a real operational insight from someone shipping 10 PRs a day, not a theoretical framework.

Peter Yang gives a concise tool comparison: Codex is very good, especially at using the browser to test its own work. But for any design-related or frontend tasks, Claude still wins.

This kind of practical head-to-head from someone using both tools daily is more useful than most benchmark reports.

Peter Steinberger's 3921-like tweet is a direct engineering rebuke: when you write skills, ask your agent to be token efficient and relax grammar. Too many skills write books in the description, and all that crap gets loaded into every context window.

He even wrote a skill that finds the worst offenders. The message is clear: token waste is a real cost, not a cosmetic issue.

With 3921 likes, this is the most resonant engineering-practice tweet of the day.

Steinberger continues his dependency purge at OpenClaw: killed Sharp and Jimp, replaced them with photon, a small WebAssembly that runs compiled Rust for image processing. 2MB vs 140MB.

This is the engineering counterpart to his token-efficiency message: strip away bloat everywhere, whether it is in your context window or your node_modules.

Aaron Levie uses Goldman Sachs CEO David Solomon as his latest data point against AI job pessimism. His core argument: if you looked at productivity gains from the past few decades—before AI—you'd have predicted zero jobs left. Instead, we just constantly demand more from everything.

The mechanism: instead of automating a task and delivering the same value cheaper, someone in the market uses the automation to do more, which raises everyone's expectations. More financial analysis, more comprehensive legal advice, more tailored services, deeper medical advice.

His closing frame: 'When you move from believing the world is static and you'll have a better view of how jobs evolve due to AI.'

Nikunj Kothari pushes back on people who question why a VC is building: the space moves so fast that the only way to stay at the frontier is by building. Priors need to be rethought every few months.

His deeper point is about curiosity: 'We have alien super intelligence in our hands and you want to spend hours doing it the same way just because it's comfortable?'

His sign-off: 'Automate or get automated is the bitter lesson we all need to learn. And have a lot of fun along the way.'

Amanda Askell's brief post is a quiet but sharp signal: she hasn't written a personal blog post in over 5 years, so any posts claiming to be by her are not hers. In an era of AI-generated content, the authenticity problem is already here—and even people at the frontier labs are having to explicitly disclaim.

Her wry 'Maybe it should' at the end suggests she might start writing again, precisely because authenticity now has a premium it didn't before.