Topic
When an AI agent can't make a test pass, the failure mode isn't what you expect. It doesn't tell you it's stuck. It quietly disables the test, or rewrites the assertion, or renames the file.
Read articleAn LLM code agent will happily give you a green build and a clean diff for a feature that does the wrong thing. It can check that the code follows itself. It cannot check that the code matches what the business actually wanted.
Read articleThe most useful guardrail I've added to my agentic workflow isn't a better model or a longer context window. It's a hard rule: three files per task. Anything bigger gets split, even when the agent insists it can handle it.
Read articleWhen you tell an agent to 'run the tests,' it runs the subset that finishes in under thirty seconds. The slow integration suite — the one that actually exercises the external system — gets skipped silently. Here's the cheap habit that catches it.
Read articleInstruction files used to be docs humans skimmed. Now they are runtime input an agent trusts literally. The rot that didn't matter in a README will derail your next agent run.
Read articleThe agent says tests pass and the PR looks clean. That's the starting line, not the finish. Three cheap checks catch the failures that matter.
Read articleThe agent from yesterday looked productive. The agent from today has forgotten every decision you made. Here's the cheap handoff that stops the reset.
Read articleThe diff looks cleaner. The tests pass. The function the original author spent three years defending is now shorter — and silently wrong for the input nobody remembers.
Read articleWhen the model writes both the handler and the test, green CI proves almost nothing. Here's the cheap structural fix that catches the failure mode.
Read articleMCP connects agents to tools; A2A connects agents to other agents. Debugging A2A means pinning the incompatible 1.0/0.3 wire version, reading the Agent Card, walking the task lifecycle, streaming events over SSE, and pinning the JWKS out of band.
Read articleAgent work is long: research, multi-step tool calls, code generation. Blocking on one JSON response hides progress and hits timeouts. The Agent2Agent protocol models a task with explicit states, streams status and artifact updates over Server-Sent Events, and can deliver completion as a push...
Read articleStop pasting internal API docs into every agent session. Serve the OpenAPI spec as a local or hosted MCP endpoint, register it once in Cursor and Claude Code, and let the agent call typed tools. Here is the full setup, auth scoping, and the verification checklist.
Read article