TIL
Short, atomic things I've learned — from events, debugging sessions, conversations, and reading.
- Claude's 'global workspace': my take on how a model thinks
Anthropic found Claude has a 'global workspace': a small set of internal patterns, like conscious thought, that it can report on, control, and reason with.
via Anthropic Research: A Global Workspace in Language Models
ai llm interpretability safety - You can't tell an improvement from a regression unless you measure both sides
Evaluating my own agent taught me that a single 'after' score is almost meaningless. The value of an eval is catching the thing you broke without realizing it.
via Evaluating Credential Sentinel
ai evaluation llm - Never tear down the old thing until the new thing has proven itself
Building a credential-rotation agent, the lesson that stuck was about ordering: keep the old credential alive as a safety net until the new one has actually verified in production.
via Building Credential Sentinel
ai-agents infrastructure security - Vibe coding is about activation energy, not replacing engineering
The real unlock of AI-assisted development isn't that it writes code for you — it's that it removes the last bit of friction standing between an idea and a first version.
via DeepStation Vibe Coding Cohort
ai vibe-coding product - Tools are only seen when the pain they solve has been felt
Watching someone encounter GitHub for the first time — and immediately understand how it would change their work — was a reminder that powerful tools stay invisible until the need exists.
via DeepStation Vibe Coding Cohort, Wynwood
developer-experience community learning