Anthropic doubles Claude Cowork usage limits for the next month. Delegate bigger, more complex tasks to Claude. 8810 likes—the highest engagement tweet of the day by far.

This is a competitive move against OpenAI's Codex, coming right after OpenAI's own reliability issues and limit resets earlier in the week.

Ben Cherny (Anthropic) describes where Cowork shines: work that's too big for a chat—research across dozens of accounts, recurring reports, triaging inbox and drafting replies. 'If you've been curious, this is a good month to find out what it can take off your plate.' 127 likes.

Thibault Sottiaux (OpenAI) observes a product maturity signal: Codex papercuts going down while adoption goes up. 976 likes.

He also notes that better memory = shorter prompts = more utility per token. 1675 likes—a concise articulation of why memory improvements compound.

Rauchg launches the Skills API: use it to make all your agents and platforms smarter. Think of it as the npm registry for agent capabilities and extensibility. Free and open. 322 likes.

This is a significant infrastructure play—if skills become the modularity unit for agents, the registry that hosts them becomes a strategic platform layer.

Rauchg announces agent filesystem state can now be read, written and mounted independently of Sandbox lifecycle. A novel virtual storage infrastructure for all compute products—storage that's decoupled but attachable to Builds, Functions, Sandboxes and more. 172 likes.

Cursor's Ryan Lu shows designing in code is now as easy as click, chat, hold shift to multi-select. Works best with Composer 2.5. 288 likes.

This is the 'yes-code' vision in practice: code isn't just for engineers anymore, it's becoming the medium for design itself.

Peter Yang's 5-step guide to building AI skills that check their own work and improve over time: 1) Give it context—provide examples of good output. 2) Make it easy to trigger—write a clear skill description. 3) Add evals—create an evals md with 10 pass/fail checks. 4) Add memory—create a memory md to capture learnings from past chats. 5) Build a skill to edit skills—create a meta-skill that cleans up other skills by removing duplicates, stale instructions, and AI slop.

This is one of the most practical and actionable guides for the skill-building era. 154 likes.

Swyx offers a smarter alternative to 'always use plan mode': always frame your task as a question, so the model is invited to push back and rate the quality of the idea, suggest alternatives, rather than blindly execute what you said to do (which is often not precisely what you meant). Literally just appending '?' to the end of your prompt often does it. 119 likes.

This is a simple but powerful prompt engineering insight: the difference between a command and a question is the difference between execution and deliberation.

Levie makes a crystal-clear argument: coding is basically the pinnacle of what you could reasonably automate with AI—the models are trained on massive code, users are highly technical, the work is verifiable (you can test an app), context is already digitized in the codebase. And yet, we still need human engineers to oversee agents.

His conclusion: if with all those advantages, engineers remain in very high demand, the risks are going to be less than perceived for other areas of knowledge work where data isn't as digitized and work isn't as easily verifiable. Agents let people do far more than before, but the people don't go away. 283 likes.

Matt Turck's satirical follow-up to the VC meme cycle: 'Bad stories about VCs are spreading on X right now, but VCs have horror stories about founders too. Like that one time a founder took another term sheet with a higher valuation despite our obvious ability to add value, thought leadership and vendor discounts.' 907 likes.