Founder @ KaaS.team, Agentic AI educator

You can build an agent in an afternoon. Designing the right system is a different job. Should code decide each step, or the model? One agent or a team? What starts it, and how long does it live? What outcome is it hired for, what must it never do, and what will each outcome cost?
Most teams answer those questions in production, one incident at a time. In this workshop you answer them on paper first, for one real job you bring. Engineers and PMs work in pairs, because a design review asks both kinds of question.
The day is six exercises, not a lecture. You take apart systems real companies run and have written about, such as SuperPlane, ClickStack AI SRE and Open Code Review, then design your own with the same canvas.
You leave with a one-page architecture you can take to a design review: the shape, the outcome and its metric, the roles and models, the approvals, the failure plan and the cost per outcome.
A hands-on design day for one real job at your company: the right shape, the outcome it owns, its approvals and its cost per outcome.
Choose a workflow, an agent or a team of agents for the job you bring.
Decide what starts it, a person, an event or a schedule.
Decide how long it lives and which steps stay with people.
Write the outcome card: outcome, metric, target and the one thing it must never do.
Define done as the system of record agreeing.
Turn each line of the card into a test the engineers can write.
Take apart a published system with the design review canvas.
Name its shape, its measure and the trade it chose.
Carry one decision into your own design.
Decide which agent does what and pick a model per role.
Mark what each agent reads, remembers and caches.
Scope each tool to the person the agent works for.
Write the approval matrix for every action that cannot be undone.
Plan the run that dies halfway, with idempotent writes.
Price the design per outcome, failed runs included.
Present a one-page architecture and answer the review questions.
Record each choice with the option you rejected.
Leave with the list of what to build first.
What's Wrong With This Design?: A support bot built as a five-agent team. Find what it costs and what breaks, then learn the map that would have caught it.
Place Your Job on the Map: Choose workflow, agent or team for your job, decide what starts it and how long it lives, and defend the choice to your pair.
Write the Outcome Card: Write the outcome, metric, target and must-never for your job. Your pair finds the activity hiding as an outcome.
In pairs, take apart SuperPlane, ClickStack AI SRE or Open Code Review with the review canvas, then report the trade it chose.
Roles, Models and Context: Draw the roles, pick a model per role, and mark what each agent reads, remembers and caches.
The Approval Matrix: List every action, mark what cannot be undone, and name who approves it and what the approver sees.
The Run That Dies Halfway: Plan retries, idempotent writes and resume for your design, then price it per outcome on one page.
Three designs reviewed live, the rest in breakouts with the review questions. Leave with what to build first.

Founder @ KaaS.team, Engineering leader, Agentic AI educator
Product managers scoping an agent feature (you decide what it owns, where a human approves and what each outcome may cost)
Software engineers about to build an agentic system at work (you choose the shape, roles and models before you write code)
Tech leads and engineering leaders who review agent designs (you leave with the questions a review asks and the answers to reject)
You have called a model from code, or shipped a feature with engineers who did.
A process at your company you want an agentic system to own.
This is a design session; you draw, decide and get feedback.

Live sessions
Learn directly from Paramanantham Harrison in a real-time, interactive format.
Live workshop access
Four hours of hands-on design with Param, six exercises on your own job.
The design canvas
The one-page architecture template, the outcome card, the approval matrix and the review questions.
Recorded design reviews
Walkthroughs of systems real companies run, to watch before and after the day.
Worked reference design
The AI mission control design from the buildcamp, with its decision records.
Feedback on your design
Live review in the room, plus written notes on designs posted within a week.
Recordings you keep
The full session, recorded.
Certificate of completion
Share what you designed with your team or on LinkedIn.
Maven Guarantee
Your purchase is backed by the Maven Guarantee.
Count input, cached and output tokens per step, then roll them up per feature and per customer.
Route simple steps to a smaller model, cache the stable prompt and trim the context the agent rereads.
A token budget per agent and per customer that stops a runaway run.
Maven for Teams
Reimbursement
Get your company to pay
Everything L&D needs: email template, receipts, and certificate of completion.
Get reimbursedTeam discount
Learn with your teammates
Save 20%+ when 2 or more teammates enroll in the same cohort.
Save 20%+ with a teamPrivate cohort
Run a cohort for your org
A dedicated cohort with a custom schedule and curriculum, tailored to your team.
Book a private cohort$350
USD
11am–5pm EST