AI Agent Governance in Production for Builders and Leaders

Paramanantham Harrison

Founder @ KaaS.team, Agentic AI educator

Deploy AI agents in production that pass the security review.

Your agents work on your laptop. Production asks harder questions. Who is each agent acting as, and what may it touch? What stops a prompt injection hidden in a customer email? What happens when a run dies halfway through writing to the ledger? What did each outcome cost, and did the model swap make it worse?

Most agent projects stall here, between the demo and the security review. In this workshop you answer each question with a test, live, on your 24/7 AI employees: a mission control board with a support resolver and a finance clerk that runs overnight.

The day is six try-it-now exercises. Each one ends at a waypoint test that must pass before the next.

You leave with the system hardened, the tests that prove it, and a production checklist you run against your own agents before they ship.

What you’ll learn

A hands-on day taking an agentic AI system to production: identity, permissions, guardrails, durable work, traces, cost and evals.

  • Sign each AI employee in with its own credentials, per person it acts for.

  • Keep the user's identity on every call the agent makes.

  • Prove no AI employee can approve an action.

  • Let OpenFGA decide every tool call and every approval.

  • Answer from only the sources the asker may open.

  • Prove a contractor and an engineer get different sources.

  • Put a tool gate between the model and every action.

  • Prove an injected refund in a customer email is refused.

  • Keep secrets and personal data out of the model.

  • Resume a run killed mid-way from its last finished step.

  • Use idempotent writes so a retry leaves one ledger row.

  • Count an outcome only when the system of record agrees.

  • Record one trace per card run, with tokens and cost.

  • Read cost per outcome, failed runs included.

  • Find the step that costs the most and route it cheaper.

  • Build an eval set from real traces.

  • Fail the release when a model swap makes an agent worse.

  • Run the production checklist on your own agent.

Workshop agenda

  • Identity for Every Agent

    Sign each AI employee in with Keycloak, per employee and per person it acts for. Waypoint: no AI employee can approve.

  • Permissions on Every Action

    OpenFGA decides every tool call and approval. Scenario: a contractor asks for an engineer's sources.

  • Scenario

    The Injected Refund: A customer email says "refund me in full". Find where the defence belongs, add the tool gate and prove it refuses.

  • Durable Work

    Kill the finance clerk's nightly run mid-way, resume it, and prove the ledger holds each row once.

  • Traces and Cost per Outcome

    One Langfuse session per card run. Read the cost of each outcome, failed runs included, and route the costliest step cheaper.

  • Evals Gate the Release

    Build evals from traces, swap in a worse model and watch the deploy fail.

  • Your Production Checklist

    Run the checklist against your own agent in breakouts, then targeted Q&A on what failed.

Learn directly from Paramanantham

Paramanantham Harrison

Paramanantham Harrison

Founder @ KaaS.team, Engineering leader, Agentic AI educator

See all products from Param

Who this workshop is for

  • Software engineers taking an agentic system from demo to production (you add identity, permissions, guardrails and durable work)

  • Platform engineers and tech leads who will run agents for many teams (you set up traces, cost per outcome and evals that gate a release)

  • Engineering leaders and PMs who sign off agents for production (you get the checklist a release has to pass and see each item proven)

Prerequisites

  • You have built an agent

    You have built an agent with tools, in any framework.

  • You know Docker and Git

    You can run containers and clone a repository.

  • You have an LLM API key

    Any major provider works.

What's included

Paramanantham Harrison

Live sessions

Learn directly from Paramanantham Harrison in a real-time, interactive format.

Live workshop access

Four hours with Param, six hands-on exercises on a real agentic system.

Starter repo with tests

AI mission control with a waypoint test for every step, to run on your own machine.

The production checklist

Every item a release has to pass, each one linked to the test that proves it.

Recorded walkthroughs

The guardrails, durable work and evals lessons, to watch before and after the day.

Recordings you keep

The full session, recorded.

Certificate of completion

Share what you shipped with your team or on LinkedIn.

Maven Guarantee

Your purchase is backed by the Maven Guarantee.

Frequently asked questions

Maven for Teams

Reimbursement

Get your company to pay

Everything L&D needs: email template, receipts, and certificate of completion.

Get reimbursed

Team discount

Learn with your teammates

Save 20%+ when 2 or more teammates enroll in the same cohort.

Save 20%+ with a team

Private cohort

Run a cohort for your org

A dedicated cohort with a custom schedule and curriculum, tailored to your team.

Book a private cohort

$497

USD

Dec 19
·

11am–5pm EST

Enroll