Your AI Coding Agent Needs a Change Budget
Why passing tests is not enough to keep autonomous code changes safe, reviewable, and reversible.
Aug 17, 20265 min read1
Command Palette
Search for a command to run...
Latest articles
Your AI Coding Agent Passed CI. Why Would You Reject the PR?
A passing test is not the same as a change your engineering team can safely merge.
Aug 15, 20265 min readYour AI Agent Said It Sent the Refund. Can You Prove It?
Why production AI agents need execution receipts, not just prompts, traces, and chat logs.
Aug 13, 20265 min read4An AI Agent’s Permission Is Not an Approval Record
How to decide which model actions may execute now, for this target, with this payload.
Aug 10, 20265 min readA Data Class Is Not a Prompt Label
Why multi-model AI products must classify the whole request before choosing a model route.
Aug 8, 20265 min readA Model Route Is Not a Permission Boundary
How to prevent primary and fallback AI routes from receiving more tool access than the workflow actually needs.
Aug 7, 20264 min read1A Fallback Model Is Not a Fallback Experience
How to decide what an AI workflow can safely keep, reduce, or stop when its preferred model route degrades.
Aug 6, 20264 min read2Your Fallback Cannot Help If You Keep Calling a Broken Route
How circuit breakers protect multi-model AI apps from degraded routes, wasted retries, and late fallbacks.
Aug 5, 20264 min read1Your AI Workflow Ran Twice. Was It a Retry or a Duplicate?
How to use idempotency keys and durable job state to prevent repeated model calls, tool runs, and side effects.
Aug 4, 20265 min read