Thibault Sottiaux (OpenAI) announces GPT-5.5 on AWS with a cheeky one-liner: 'Heard that AWS is where the cool kids are. Hello. We have GPT-5.5.' 835 likes.

Levie adds context: AWS has massive enterprise traction with large committed contracts. This partnership opens up increased distribution for OpenAI's models and likely drives an increase in token consumption overall across model providers.

Rauchg reports MiniMax M3 is now the leading open model on the Next.js agent evaluations—right behind Opus and GPT-5, but 10x cheaper (and 20x cheaper on Vercel AI Gateway). 596 likes.

This is a significant signal: open models are closing the gap with proprietary ones on real-world agent benchmarks, not just academic tests.

Peter Yang distills Josh Shpigford's top 6 lessons for building multiple products solo with AI agents. The most practical: have GPT review Claude's work and vice versa ('GPT invariably finds 3-5 bugs that Opus overlooked'). Different models spot different mistakes.

Second key insight: build a /learnings skill that runs at the end of each phase so every 'no, that didn't work' moment gets distilled into new rules. And use git worktrees to build features in parallel without context rot.

The humility note: Josh credits 25 years of building before AI for how fast he ships now. His advice for newer builders is to 'just fail a lot, because the only way to figure out what not to do is by doing the thing incorrectly.'

Dan Shipper's one-liner captures the new work ethic: you don't have to work 7 days a week if you just have a swarm of Codex's running on /goal all the time. But you'll probably want to. 248 likes.

Steipete tells Codex to use voice whenever he's distracted and it needs his help to be unblocked. Every once in a while he hears it talking to him, and says it's the coolest thing ever—for example, for releases that need npm and are 1Password-gated.

This is a new pattern in agent-human interaction: the agent initiates communication when blocked, rather than waiting for the human to check in. 683 likes.

TRQ shares how Anthropic employees stay in the loop with Claude: he's been asking others at Anthropic how they fully understand the work being done. This is one of his favorites from Suzanne, who uses voice mode to make it easier and more natural to respond.

6183 likes makes this the highest-engagement tweet of the day. The signal: even AI company employees are figuring out how to work with their own products in daily practice.

Levie addresses the defining question of the AI era: how do you develop competitive advantage when your competitor has access to the same AI models? His answer: the companies that best harness their internal institutional knowledge, existing data assets, and domain-specific workflows—connected with AI—will stay ahead.

The key is to capture and protect the value created by unique data, processes, and expertise. Each industry will have its own version of this. At Box, customers want to take advantage of their institutional knowledge and have the flexibility to bring any AI model to their data at any time.

Matt Turck's CEO vs CTO meme: CEO says 'we have tens of thousands of AI agents running in production at massive scale right now.' CTO's reaction is silent panic. 289 likes.

This captures the gap between narrative and reality in enterprise AI deployment that Levie has been discussing all week.