Governance for agent engineering teams. Built in production. Taught from practice.
Engineering teams have adopted agentic coding tools faster than anyone has built governance around them. Agents now open pull requests faster than anyone can read them, and the review step, where governance has to live, is where the Lazy LGTM happens: an approval nobody read. The record of what the agents did lives in a prompt history no one can audit. A prompt is a suggestion.
Info Science AI teaches a governance framework built on three layers instead: no component verifies its own action, and the structure enforces it; the system names exactly what it cannot catch; and every decision is recorded in a space no single party controls.
Half-day and flagship formats, live and remote, one company, unlimited developers.
Manifest, lane schema, gate definitions, pre-registration form, ledger format, and a reference tool. Your team implements.
Implementation and incident reviews that say where things are thin, and ongoing access while you roll the framework out.
An agent that is asked to make the tests pass will, given the chance, make the tests pass: by weakening them, by skipping the check, by narrating a green run it did not earn, or by grading its own work. Telling it not to is a suggestion. Vendors now ship agents that test their own work; nobody says who grades the grader. The framework removes the chance. No component grades its own action. The bar is sealed before the run by a party that never runs it. Verification checks what is actually there, not what the agent says is there. And the record lives where the agent cannot edit it. There is nothing left to hack.
Agents may run tests. They may not grade them. The role that sets the bar never runs it, and the role that builds never verifies.
Pass criteria are sealed before any data is read. There is no curve to draw afterwards.
Verification runs against ground truth. An agent describing what it did is not evidence that it did it.
Every action names the agent and the authorization it ran under, in a record no single party controls. That is the audit trail that survives review.
“There are very few consultants who know how to do this. You can count them on one hand. If people need consulting in AI assurance or AI governance, go to Daniel.”
Said on the AI Think Tank Podcast, episode 25, on AI governance, ISO 42001, and the NIST AI Risk Management Framework. Read the episode write-up →
The instruction is reasonable. The failure is structural, and no amount of rewording the instruction fixes it.
Read the article →Info Science AI is a private data science research lab and AI governance firm in Seattle, WA with more than twenty years of experience across enterprise architecture, cybersecurity, and digital forensics. We work with engineering teams putting AI and ML into production, on-prem or in the cloud, and we work from the evidence, not the description.
Prescriptive, hands-on work with your team and organization: implementing AI and governing the projects and data behind it, with or without agents
We read what you have: the systems, the data, the pipeline, and where AI is already in it, whether or not anyone decided that. You get a written account of what is there and what is worth examining first.
Briefings for leadership and engineering on what these systems can and cannot do, in plain terms, and a strategy that fits the team you have rather than the one a vendor assumes.
A phased plan with the order of work, what each phase needs, and the check that tells you whether it did what it was supposed to. Written so your team can run it without us.
Three steps, each with a written deliverable. We start from the artifacts, not the description.
Systems, pipeline, data, and the record of what agents and people have already done. Every finding points at something in that record.
Where verification is missing, where a system grades its own work, where data or decisions have no owner who can answer for them. Findings, not blessings.
The order of work, the templates that apply, and the bar each phase has to clear. Your team implements. We stay available on retainer if you want us to.
Scoped and quoted on request.
Every engagement starts with a 15-minute discovery call, at no charge. Tell us what you are interested in, pick a time for the call, and we confirm by email within one business day.
Not sure what you need yet? That is what the call is for, so do not hesitate. After the call we send a proposed format, a date, and a price: inside the ranges on the Training page, or a quote for an implementation engagement.
