Notes
Guides, lessons learned, open topics and announcements. The method behind the work, written down.
- Clean up after your AI agentsAI agents leave a working copy of the code behind for every task, and nothing removes it. The three checks we use to remove one safely, and why cleanup is part of done.
- We built an AI dispatcher, then bypassed it ourselvesWhether a team uses an AI agent depends less on how good the agent is than on what it costs to hand it a task. What we saw on day one, and what we changed.
- Keeping AI agents under control: a method for predictable agent workHow to keep AI agents' work safe and predictable: people start the work, tasks a machine can check, a budget for each task, guards the agent cannot bypass, and a review.
- Data ingestion at scale: a metadata-driven framework for hundreds of sourcesHow to ingest hundreds of data sources with one metadata-driven framework: source records, standard loaders, a layered lake, tested models and onboarding metrics.
- Private AI chat for a company: what it is, when it pays, and the decisions that matterHow to give a company its own AI chat: hosting, models, a gateway for keys and budgets, directory login, knowledge bases, and a rollout your IT team can run.
- Reporting automation: cost, architecture and a worked exampleHow to price, design and justify reporting automation: the manual-cost formula, the four layers of a pipeline, payback, and a labelled hypothetical example.
- Ask the system what made the work harder, and let a person decideA small retrospective after every working session and every AI agent task, which proposes changes and never makes them. How ours works, and why it stays useful.
- A guard that does nothing looks like a guard with nothing to doAutomated guards around AI agents fail silently more often than loudly. How to test that a guard fires, that it stays quiet, and that it says what it cannot see.
Book a 30-min scoping call
Bring one process. We will tell you whether automation changes it, and what it would take.