Blog
Notes from the team building the agentic frontend: product, engineering, company.
Documentation drift in AI assistants: when a stale source becomes a confident answer
Your AI assistant says exports cap at 10,000 rows. An engineer raised it to 50,000 three sprints ago. Here is how documentation drift turns an old source into a new wrong answer, and what can catch it.
28 Oct 2026Human-in-the-loop approvals for AI agents: the approval that arrives on Monday
Your agent asks for a yes on Thursday and gets it Monday, after two deploys and a policy change. Here is the architecture that makes waiting for a human boring.
24 Oct 2026AI agent permissions: allow, ask or deny, and the yes budget
Which agent actions should just run, which should wait for a yes, and which should never exist? This post covers allow, ask and deny, whose keys the agent holds, and how many approvals your users will tolerate.
10 Oct 2026How to add an AI copilot to your SaaS product: the iceberg under the chatbot
A copilot demo takes a weekend. Shipping one takes identity, permissions, durable turns, fresh knowledge, model fallbacks and a real UI. Here's the whole iceberg.
26 Sep 2026The agentic frontend: the end of the interface tax
Your users don't want your UI — they want what's behind it. We've spent forty years pointing at rectangles and paying the interface tax, and the next frontend of your product finally ends both.
8 Aug 2026