Your engineers have read about agents. They have not shipped one.
Hands-on AI training built around your stack, your codebase and your constraints — taught by the engineers who build and operate these systems in production, not by a trainer reading slides.
Your codebase. Your engineers. Code they keep.
Why most AI training changes nothing
Your team comes back energised on Friday and is back to normal by Wednesday. Four reasons, and we built the workshops to avoid all four.
Somebody else's example
A chatbot over a sample dataset teaches nothing about your authentication, your data model or your deployment path.
No production path
The notebook never becomes a service. Nobody shows the team what the deployed, monitored version looks like.
Security left out
Nobody shows the team how an agent gets abused, so they build the version your reviewers will reject.
Nothing survives the week
The sandbox is torn down on Friday. On Monday there is no repository, no runbook, nothing to build on.
A course, or a week of shipping
We are not accredited to deliver official curriculum, and we do not pretend to be. We teach the only way we know how.
- A fixed syllabus, the same for every client.
- A trainer who has not operated the thing being taught.
- Exercises in a sandbox that gets deleted.
- A certificate, and no change in your repository.
- The syllabus is written after we read your stack.
- Taught by engineers who run production AI systems.
- Your repository, your data, your deployment path.
- A working agent, a test suite and a runbook you keep.
The three-day core, day by day
Three consecutive days with your engineers. We write the agenda after a scoping call, against your stack.
Three tracks, by who is in the room
The engineering track is the main one. The other two exist because engineers cannot ship what the organisation will not approve.
The three-day core. Your engineers build one agent against your own systems, attack it, and harden it. They leave with the repository, the tests and the runbook.
What to ask for, what to refuse, and how agent risk differs from application risk. We demonstrate the attacks live, then write your review checklist together.
Where AI spend goes to die, how to tell a real pilot from theatre, and what a production agent costs to run. Short, direct, and free of vendor optimism.
What your team leaves with
Four things, and none of them is a certificate.
A working agent, in your repository
Built by your engineers against your systems, with tests. Not a sandbox that disappears on Friday.
The attacks, run by hand
Your team breaks its own agent, so it never again ships one that your reviewers will reject.
Your own patterns, written down
The conventions the cohort agreed — permissions, logging, evaluation, escalation — as a document your next project starts from.
A review checklist your security team wrote
So the second agent does not spend three months in approval the way the first one did.
How we run it
Three rules that make the week work.
Small cohorts
Six to ten engineers. Beyond that, people stop typing and start watching, and watching teaches nothing.
Your code, not ours
We teach against your repository and your data. Everything the cohort writes stays in your repository, under your licence.
On site or remote
On site works better for the attack day. Remote works, in your timezone, if travel is not practical.
Taught by the people who operate it
tMinus1 built Automatum — multi-tenant SaaS, metered billing, marketplace integrations across AWS, Azure and GCP. 80+ software companies run their marketplace revenue through it, and we still operate it. The examples in the room are systems we have been paged about.
No, and we will not imply otherwise. We are an AWS Advanced Tier Services Partner, not an accredited training provider, so there is no official curriculum and no AWS certificate at the end. If you need accredited courses, we will point you at them.
They should be able to write and ship code in your stack. No machine learning background is needed. If your team can build and deploy a service, they can do this week.
Not for the three-day core. We teach against a non-production environment and anonymised or synthetic data where that is what you prefer. The optional fourth day is the only part that touches a real environment, and only through your change process.
Keep the build days to engineers. The attack demonstration on day 3 is worth opening to your security team, and the leadership session works better separately and shorter.
One fee per cohort, quoted after the scoping call, because the agenda depends on your stack and the process you pick. Travel is quoted separately and at cost.
Three days. One agent. Your code.
Tell us your stack and who is in the room. We will tell you what the week would cover.
