AGENT OPERATIONS COURSE // WAITLIST STATUS: COHORT 1 FORMING

Practitioner course // name pending

Run AI agents you can PROVE are telling the truth.

Most people are taught to build agents. This is about running them: a governance layer over every agent you operate, and receipts for what each one actually did, what it cost, and whether its output checked out. If it cannot show its work, you do not ship it.

42

Lessons written

SRC: lessons/ directory count, 2026-08

3

Public proof repos

SRC: linked repository pages, checked logged-out 2026-08-26

65

Mining batches of real incidents

SRC: BRAND-PLAN.md, 2026-08-25
FILE 01 / EVIDENCE

Why listen to this lab and not another course seller

  • EXHIBIT A
    The work is public. The builds behind this course live in open repos: gpu-cpu-mutex, agent-time-ledger, and TreeTrace. Read them before you join. Nothing here asks you to trust a screenshot.
    Source: the three repository links above, checked logged-out 2026-08-26
  • EXHIBIT B
    We publish our own failures. A backup job here once reported success while writing four empty archive files. Nobody noticed for days. That incident became a standing rule of this lab: an exit code of zero is not evidence. Every lesson starts from mistakes like that one, not from a highlight reel.
    Documented in this lab's own postmortems
  • EXHIBIT C
    Costs get published, not hidden. Every build taught in the Lab ships with its token and time cost totals published alongside it. You will know what an agent fleet costs to run before you bet your own money on one.
    Standing commitment of the Lab
“Yall never show proof just farming clicks.” A comment we built this whole offer to answer
“Who pays for all the LLM tokens? … I will sign up today.” The most requested missing number in agent education
FILE 02 / OUTCOMES

What you leave able to do

01

Catch a lying process

An agent that exits cleanly can still have done nothing, or done the wrong thing. You will stop treating a green status light as proof, and start checking the artifact itself.

Verification
02

Verify where it matters

You will test outcomes the way the person depending on your agent experiences them, not the way the logs happen to look when the pipeline passed internally.

User layer
03

Break your own checks first

Before trusting any verification that passed, you will run it against input known to be bad. A check that cannot fail tells you nothing. You will make yours able to fail.

Negative control
FILE 03 / ENROLLMENT

Get the invite

The first cohort is small and opens by invitation. Joining the list costs nothing and commits you to nothing.

Founding rate locked

Founding member rate

The first cohort locks its membership rate permanently. When cohort two opens at a higher price, founding members stay where they started. There is no countdown timer on this page because there is no fake deadline: the rate changes only when a real cohort has come and gone.

Cohort 1 pricing never repeats. That is the whole mechanism.