Skip to content

Cognithamus Institute

How engineering organizations operate when AI agents become part of the team.

Not how to use AI. What changes underneath an engineering organization once a meaningful share of its work is produced by something that is not a person — how work is delegated, how output is reviewed, how permission is granted, and how any of it is measured afterwards.

The view

  • 01

    A claim without a measurement is not a result.

    Most reporting on agent productivity measures motion — commits, lines, tickets closed — because motion is the easiest thing to count and the first thing an agent produces more of. The question that survives scrutiny is narrower: what shipped, and would it have shipped anyway.

  • 02

    Agent output is reviewed, not trusted.

    Review was a check on a colleague who understood the system. It is now the primary control on work produced by something that does not. That is a different skill, applied at a volume no team was staffed for, and almost nowhere is it taught.

  • 03

    An agent's permissions are an org chart.

    What an agent may read, change and merge decides what it is accountable for and who answers when it is wrong. Most organizations have granted that structure without designing it, by configuring a tool.

What we study

Five areas, each an open question rather than a course.

  • The Review Discipline

    Delegation, and reading work produced by something that cannot explain itself.

    What does a reviewer need to see to accept a change they did not write and cannot ask about?

  • Software Archaeology

    Understanding an estate no living employee wrote.

    How much of a system's intent can be recovered from what it left behind — its history, its structure, its scars?

  • Agent Governance

    Permission, attribution, and who answers for an agent's work.

    Where does accountability sit when the author is not a person?

  • Evidence and Evaluation

    Judging agent performance without buying the theater.

    Which measurements distinguish an organization that got faster from one that got busier?

  • Adoption Under Uncertainty

    Introducing agents as a change to process, not to tooling.

    What breaks first, and what should a team stop measuring before it starts?

The one paid program

AOS — Agentic Operating System

A personal operating model for individuals redesigning how they work alongside agents, drawing on the areas above. Module 00 is published in full and free to read. Everything else in the Institute is free and stays free.

Read about AOS

Work with Cognithamus

Engineering Reality Assessment

A short engagement run against your own repositories rather than a slide deck: what the estate actually contains, where the work is going, and which of your current measurements would not survive being checked. The output is a written assessment you keep. It is delivered directly, one organization at a time, which is the only scale at which it can honestly be offered.

Cognithamus@proton.me

And then

Learn the principles here. Measure your own engineering reality with XI.

Every position on this page turns into a question about your own organization, and none of them can be answered from the outside. XI answers them from your repositories and your history — the same rule applied to your estate instead of the idea of one.

What this is not

  • No certificates. Nothing here is credentialed, and a certificate from a company of this size would be worth what it cost to print.
  • No accreditation. The Institute is not a school and does not claim standing with any body that governs them.
  • No cohort, and no enrolment. There is no intake, no start date, and no seat to reserve.
  • No curriculum. The five areas above are subjects under study. When written material exists it will appear here, and until then this page does not imply that it has.