Two days that change what one engineer can ship.
A two-day, in-person course that takes an engineer from prompting AI to orchestrating it. Run on your own repository and your own backlog, so you leave with the harness installed, real pull requests merged by a human who read them, and an internal agent you built and proved.
Not a course about AI. A step-change in what you can ship.
Most AI training teaches the concepts and sends you home with a reading list. This runs on your repository, your ticket system and your backlog. Every module ends in something committed: the rules file every AI tool reads before touching your code, the hooks that stop model output reaching a branch unreviewed, a protected branch, a merged pull request. Any engineer can come: this is sold per seat, and cohorts are deliberately mixed across companies.
Any engineer shipping real product
Come on your own, or with two or three from your team. You need a repository, a ticket system and the right to merge. That is the whole entry requirement.
Engineers prompting, not orchestrating
You use AI as a faster autocomplete today. You want to hand an agentic loop a real ticket and review what comes back instead.
Engineering leaders chasing output without chaos
You want the multiple on delivery, with the review discipline, the gates and the audit trail in place before the throughput arrives.
Two days. Both of them on your own work.
Timestamped, and the same every cohort. Homework between the days is one more real ticket through the harness you built on day one.
The working pattern
Ad hoc prompting versus an agentic harness. A live demo of the full loop: ticket in, plan proposed and approved, implementation with tests, multi-agent review, a named human who approves and merges, and the learning written back for next time.
Install the harness
Hands on keyboards, in your own repo. Write the rules file every AI tool reads before touching your code: coding standards, review rules, what AI may and may not touch, secrets kept out of every prompt. Wire one connection to your own ticket system. Set the first custom command for the ticket-to-merge lifecycle.
Hooks and gates
What actually stops model output reaching a branch unreviewed. Pre-commit and pre-push hooks: lint, test, secrets scan. Branch protection with required reviewers on the real repo. The rule: the model works, a named human always approves and merges, never the reverse.
Lunch
Provided both days.
The first change
Every engineer picks a real ticket from their own backlog and runs it through the harness end to end: brainstorm and record the decision, plan, implement with tests against the rules, review, human gate, merge.
Reviewing AI work
Trust boundaries. What to always check by hand, what never merges blind, and how to review a pull request a tool helped write, without rubber-stamping it or redoing the work yourself.
Retro and homework
What shipped, what got stuck, what surprised people. One more real ticket through the same harness before day two.
Homework review
What shipped overnight, what got stuck, blockers cleared as a group.
Agent anatomy
The loop, tools, and context. Concepts first, so the pattern outlives any vendor decision. A one-page translation map, mapping every technique to whatever platform you actually run.
Lab: build the agent
In pairs, against one of your real internal systems. The harness from day one applied to agent-authored work: the same rules file, the same review discipline, extended to a fan-out of specialised review passes (architecture, security, performance, migration, test coverage, code quality), with the highest-severity finding blocking merge and one named human accountable for the decision.
Lunch
Provided both days.
The quality gate
Implemented, evaluated, and proven, as three distinct states, in that order. Build an eval set and a pass threshold for the agent from the morning lab, and run it live.
Cost and model churn
What this costs to run at scale, and how the harness and eval suite survive a model upgrade without a rebuild.
The team brain
Everything from the two days, conventions file, skills, decisions and quality gates, assembled into one governed repo with a named owner and a weekly review routine. Graduation.
Six artefacts, all of them in your environment.
Your team conventions file
Committed to your own repo: coding standards, review rules, what AI may and may not touch, secrets kept out of every prompt.
Two merged pull requests per engineer
AI-assisted, against real backlog tickets, human-reviewed and shipped. Not a demo branch. Work that is in your product.
A working internal agent
Built in pairs against one of your own systems, with a fan-out of specialised review passes behind it: architecture, security, performance, migration, coverage.
An eval suite and a pass threshold
Implemented, evaluated and proven, as three distinct states in that order. You leave with the quality gate report showing pass against threshold.
A platform translation map
One page mapping every pattern taught to whatever you run, so a model change or a vendor decision never costs you a rebuild.
A team brain repo
Conventions, skills, decisions and quality gates in one governed repo, with a named owner and a standing weekly review.
The engineer who runs this every week.
“Two days in, the harness is in your own repo, there are pull requests merged by a human who actually read them, and an agent you built with an eval suite attached. That isn’t a faster autocomplete. It is a different way of working, and it stays after we leave.”
Sean has been working with AI and machine learning since long before the current wave, and is one of the architects of how possibl builds and ships production AI every day. This course is the harness his own engineers run: agentic loops handed real tickets, specialised review passes, gates that hold.
Everything except the reading list.
Included in the course fee. The community of practice below is a separate, optional plan.
Both days, fully facilitated
In person, on your own repo, with a possibl engineer beside you when it breaks.
Lunch and refreshments
Provided both days. Venue confirmed with your place.
Your artefacts, kept
The conventions file, translation map, eval suite and team brain repo all stay in your environment.
Pre-work and a readiness check
Before day one we confirm your repo access, ticket system and merge rights, and send the setup steps, so the room starts building at 9:00 rather than troubleshooting.
The community of practice.
The harness keeps moving after you leave: models change, tooling changes, and the practices we teach get sharper every cohort. The community of practice is where that keeps reaching you. It is a separate monthly plan, not part of the course fee. Add it when you apply, or later.
A place to actually ask
Discuss with other engineers running the same harness, and with the possibl engineers who teach it. Real problems, real repos, no vendor forum theatre.
Everything we publish
Access to the documents, playbooks and rules files we write as we run this in production, yours to lift straight into your repo.
New modules as they land
When a technique earns its place, it goes into the community first: new commands, new review passes, new eval patterns.
Hosted on rcrt
It runs on our own platform, so over time you get runnable modules, not just reading material.
Leave with it running, or don’t pay.
If you finish day two without the harness running in your own repository and at least one AI-assisted pull request merged by a human on your team, we refund the course in full. That is the whole point of running it on your work instead of ours.
The things engineering leaders ask first.
Who runs the two days?
Sean Muller, possibl’s CTO and co-founder, joined by a second possibl engineer as the room grows, so nobody spends the day waiting for help.
Do I need to be a senior engineer?
No. You need a repository you can clone, a ticket system, and the right to merge. Everything else we teach in the room.
Which platform do you teach?
The loop, the tools and the context first: concepts that outlive any vendor decision. You leave with a one-page translation map from every technique to whatever you actually run: Claude, Copilot, Gemini or rcrt.
Are the two days consecutive?
Yes. Both days in person, back to back, with one real ticket run through the harness as homework overnight.
Will AI be merging our code?
Never. The model works; a named human always approves and merges. Hooks, pre-push checks and branch protection go in on day one precisely so that stays true as throughput rises.
Can several engineers from one company attend?
Yes. Places are sold per engineer, so come alone or bring the team. Cohorts are deliberately mixed, and teams often send two or three.
Is the community of practice included?
No. It is a separate $49/month plan on top of the course fee. The two days stand on their own; the community is for the engineers who want the practices, playbooks and new modules to keep arriving after they leave.
What if the cohort does not run?
The Lab runs on a minimum cohort size. If we do not reach it, or the dates have to move, we tell you at least 10 working days before day one and you choose a full refund or a transfer to the next cohort. Nothing is charged until your place is confirmed.
Can I change or cancel my place?
Yes. Cancel more than 10 working days before day one for a full refund. Inside 10 working days the fee is transferable to a colleague or to the next cohort, but not refundable. Send it to hello@possibl.ai and we will sort it.
Can I join remotely?
Not for this cohort. The work happens on your machine in the room, with someone next to you when the harness fights back.
Two days, and your team ships differently.
Wellington, 22–23 September 2026. Places are confirmed after a short readiness check (repo access, a ticket system, merge rights), so nobody spends day one blocked.
The same two days, run in Auckland.
Auckland cohort, dates to be confirmed
Identical course, Auckland venue. Register interest and we will confirm the dates as soon as numbers allow.
- Same two days, same six artefacts
- Individual engineers and whole teams both welcome
- A Claude day for founders and business leaders is next