# OneDroid Blog

- 2026-10-21 [50,000 lines overnight: how do you know what it did? Hold the code to what it promised](/blog/case-study-code-vs-user-guide.md): A case study for product owners and leaders. A decades-old Fortran library held to its own user guide with 100 checks, sealed before the run and kept from whoever builds. 91 held, 9 failed, every check shown able to fail, and the test boundary published. The method that answers "how do I validate what an AI agent built?".
- 2026-10-20 [Before you add a separate tester to your AI coding agent: what one experiment measured](/blog/case-study-claude-code-alone-vs-tester.md): A case study for product owners and leaders. The same feature built six times, three by Claude Code checking its own work and three by Claude Code with an independent tester, judged on checks neither had seen. The quality was the same. The agent got cheaper and the whole got dearer. What it cost, what it showed, and what it does not prove.
- 2026-10-20 [Before you modernize legacy code with an AI agent: what the tester checks decides whether it helps](/blog/case-study-modernization-three-setups.md): A case study for product owners and leaders. A decades-old Fortran library ported to Go by Claude Code in three setups and judged on the same hidden checks against the original. A tester pointed at the manual made the port worse; a tester pointed at the original matched the agent alone at 41% less cost. What it cost, what it showed, and what it does not prove.
- 2026-10-07 [Before you put an AI agent in front of trading data: what one test found in a FINOS CALM demo](/blog/case-study-finos-calm.md): A case study for product owners and leaders. Seven checks, sealed before the run, asked whether an open source demo that puts an AI agent's tool server in front of trading data does what its architecture says. Its guardrail held in search but not on a second path to the same data. What it took, what it caught, and what it does not prove.
- 2026-10-07 [Before you move to a new release: what one test found in OTRS and Znuny](/blog/case-study-otrs-znuny.md): A case study for product owners and leaders. One sealed set of checks ran against OTRS Community Edition 6.0.30 and its fork Znuny 7.3.7 and found six changes that other systems rely on, including a fault in the old release that a plain comparison would have called a regression. What it cost, what it caught, and what it does not prove.
- 2026-10-07 [How to make sure a merged fix is actually live: 6 steps](/blog/how-to-check-a-merged-fix-is-live.md): Six steps for checking that a fix your team or your AI coding agent merged has reached the page people read. They come from three slips on our own sites, the longest of them 8 hours. Includes what the check does not prove.
- 2026-09-30 [How to configure your MCP servers once for Claude Code, Claude.ai and ChatGPT](/blog/how-to-configure-mcp-servers-once-for-every-ai-app.md): Eight steps to set up your MCP tools and credentials in one place and reach them from Claude Code, Claude Desktop and claude.ai, with the same setup offered for ChatGPT, Cursor and VS Code. Includes a check to rerun whenever a client stops seeing them.
- 2026-09-30 [How to be in control of every MCP call your AI agent makes](/blog/how-to-control-every-mcp-call-your-agent-makes.md): Ten steps to see and control every tool call Claude Code or any MCP client makes: one gateway, one log of who called which tool and when, and a switch to turn a tool off. Ends with a test call you find in the log yourself.
- 2026-09-30 [How to give your AI agent memory you fully control, with no vendor lock-in](/blog/how-to-give-your-ai-agent-memory-you-control.md): Seven steps to give Claude Code or any MCP client a memory that outlives the session, stored in your own Postgres or Supabase if you want it there. Ends with a test where a new session finds what an earlier one wrote.
- 2026-09-30 [How to test code your AI agent wrote: 5 steps](/blog/how-to-test-code-your-ai-agent-wrote.md): Five steps for testing code a coding agent wrote, using test cases the agent never sees. They come from the day this method found two bugs in our own build while every visible test was green. Includes what the method does not prove.
- 2026-09-30 [5 best open-source repos for running agentic coding loops](/blog/open-source-repos-for-agentic-loops.md): Five open-source repositories you can run today for agentic coding loops, each covering a different angle, with licence, stars, last push and one documented weakness per project. One of them is OneDroid's own.
