The conversation model gives you system, user, and assistant roles. None of them are designed for async events. Wrote about the pattern we use to handle this with AI SDK. gdaybrice.com/the-missing-me…

We kept watching our agent rediscover the same things case after case. Find the right contact channel, negotiate a good outcome, then forget everything by the next one. So we built a post-mortem agent that remembers. gdaybrice.com/the-post-morte…

If you're building tools for AI agents, you don't need MCP. A good CLI with solid docs is all they need. ejholmes.github.io/2026/02/28/mcp…

💯

Google gets it. Their new workspace tool is cli-first, mcp second. gws mcp just wraps the cli underneath. github.com/googleworkspac…

still can't connect the same GitHub account to multiple Vercel teams 🫠

x.com/i/article/2036…

Wrote about the triage agent we built at brand.ai. It checks seven observability sources in parallel, clusters related issues, and writes a full report. The hard part was stopping it from stitching unrelated facts into confident sounding narratives.

linemark now works with 40+ AI agents. Review your agent's changes locally, highlight code, comment, submit feedback straight back to the agent. github.com/gdaybrice/line…

Shared thoughts about the disposable UI pattern a few weeks ago, x.com/gdaybrice/stat…