AI Product Management, taught on your own feature

A course for people who own the AI feature

Most AI courses teach you to use a tool. This programme teaches you to decide what an AI feature should do and to prove it works before you ship. You bring one real feature, and every session moves it forward.

Marcus Chia designs and ships AI products at Produlogi, a UX × AI studio in Kuala Lumpur. The course carries two habits from that client work. The spec comes before any prompt, and evals start with reading what users typed.

Format

Primary: 4-week live cohort

• Eight live online sessions over four weeks, two a week. Session length: TBC.
• Sessions run in GMT+8 hours, with recordings for anyone who joins from another timezone.
• Each session ends with an artefact you apply to your own feature before the next one.
• Written feedback from Marcus on your capstone.
• TBC: start date, cohort size and session time.

Alternative: 2-day in-house intensive for teams

For a product team that wants to set up its AI practice together. We compress the modules into two days and use one of your features as the shared case, so the team leaves with an agreed PRD and eval plan for it. Online or on-site: TBC.

Modules

Each module has a short input, a worked example and a hands-on block where you produce the artefact.

1. AI product sense and use-case selection. Where AI earns its cost and where a rule or a form works better. You score candidate use cases on value and risk, then choose between a single model call, a workflow and an agent. Artefact: a scored use-case shortlist with your pick.
2. Designing AI UX. How much the AI may do on its own, and how people see its uncertainty and undo its mistakes. You map the states a demo skips: loading, low confidence, wrong answer, refusal and recovery. Artefact: an autonomy decision and a state map for your feature.
3. Context, prompt, harness and loop engineering for PMs. What the model sees at each step and where a human stays in the loop. You learn enough to make the product calls and talk to engineers in their terms. Artefact: a one-page context and harness spec.
4. Evals and error analysis. Read real traces, code the failures the way a researcher codes interviews, then turn the top failure modes into binary pass/fail checks. You write one LLM-as-judge check and measure how often it agrees with human labels before you trust it. Artefact: an error taxonomy and a first eval set.
5. Observability and shipping. Traces and guardrails, then a rollout plan with the metrics to watch and a rollback trigger. Artefact: a launch checklist and monitoring plan.
6. The AI PRD and roadmap. A PRD that states what "good" means as eval criteria, plus the cost, latency and failure budgets. A roadmap that sequences learning before scale. Artefact: the first full draft of your AI PRD.
7. Working with engineers and the FDE model. How to split decisions between PM, designer and engineer, how to review an agent's work, and when a forward-deployed partner beats a hire. Artefact: a RACI for your feature and a review ritual your team can run weekly.
8. Capstone reviews. Each participant presents their feature and gets critique from Marcus and the cohort.

Capstone

You finish with three connected pieces for one real feature:

• An AI PRD that defines the job, the user, the autonomy level and the quality bar.
• An eval plan built from your own error analysis, with a validated judge and the thresholds that decide ship or hold.
• A working prototype that shows the core flow, including at least one failure state and how the user recovers.

Keep them. They are portfolio pieces if you are changing roles, and a starting brief if your team is about to build.

What you leave with

• The ability to say whether a feature needs AI at all, and which kind.
• A repeatable way to read traces and turn what you find into evals.
• Interface patterns for uncertainty and correction that you can hand to a designer or build yourself.
• A shared vocabulary with engineers for context, harness and runtime decisions.

Who it's for

• PMs moving into AI product roles, or asked to own an AI feature for the first time.
• Designers who want to own the behaviour of AI features, beyond the screens around them.
• Founders scoping their first AI product and deciding what to build before they hire.
• Product leaders setting up an AI team and its review practice. The team intensive suits you best.

Who it's not for

• ML engineers looking for model training, fine-tuning or infrastructure depth.
• Anyone who wants a certificate to add to a profile without applying the work to a real feature.
• Teams that need an HRDC claim. The AiTraining2U programmes cover that route.

How it compares

US-led AI PM cohorts run three to six weeks in US hours. Most end with a project on a sample product. We made three different calls:

• UX × AI first. Autonomy, uncertainty and error recovery get a full module. Most AI PM courses give them a slide.
• Research-backed evals. Error analysis runs as qualitative research on traces, and judges get checked against human labels. Read the thinking in LLM error analysis is user research.
• Your feature, your artefacts. The capstone uses the feature you bring, so the work carries straight into your job.

Sessions sit in GMT+8 hours, which works for Southeast Asia and much of the Asia-Pacific without a 2am call.

Your trainer

Marcus Chia runs Produlogi and leads this programme. He has a long UX and UI background, and he tests, builds and ships AI features as part of client work. He also teaches as faculty at AiTraining2U, where he is the named instructor for AI Vibe Coding and Rapid Prototyping. Read more on the about page.

For the background reading behind the modules, start with Harness engineering: a PM's guide to the stack and AI product manager skills and the FDE hype. The free AI Feature PRD Template follows the same PRD structure as the course.

Status

This is a new programme. TBC: first cohort dates, cohort size and session time. There are no past cohorts yet, so this page shows no participant numbers or testimonials. Register your interest and tell us whether you want a seat in the cohort or a private run for your team.