Enterprise learning · Product companies
Train your engineering team to ship AI agent features that hold up in production.
Live, project-based learning with Sunil Mathew for the engineers, architects and engineering leaders who build and own your product's AI features.
The demo works. The harder questions come next.
Your roadmap has AI features on it, and the first version probably works in a demo. What the team has to answer next is harder: what the agent may do on a customer's behalf, how you will know when it gets something wrong, what each run costs at your traffic, and how you change a model or a prompt without breaking a release.
Those are engineering decisions, and they sit with the people who build and own the product. The learning gives them practice making those decisions on a working system, with Sunil reviewing the reasoning.
An example the group works through
Illustrative. An in-product assistant that can change a customer's subscription. The group builds a version of it, then tests what happens when the same request arrives twice, when the customer's permission is missing, and when a cheaper model changes the answer.
Each case ends in a decision the team can explain in its next design review: what the agent may do alone, what needs a person, what the evaluation checks before release, and what a run may cost.
You can start from an industry example. Confidential material is not needed for a first conversation; company-specific work, data access and any agreements are scoped separately.
Who takes part
- Engineers
build the workflow, test its failures and measure behaviour and cost.
- Architects and tech leads
set boundaries, review trade-offs and decide what evidence a release needs.
- Engineering managers and directors
guide the review and decide who owns approvals, exceptions and rollback.
Product and programme managers can join where the scope includes how a feature's limits are explained to customers.
Led by Sunil Mathew
Sunil has 26 years of engineering experience, including past leadership roles at Google, Amazon and Walmart Labs. He has hired, developed and assessed senior technical professionals, and today builds agentic-AI products alongside consulting and teaching.
His current work is building agentic-AI products, so the practice reflects the decisions product teams are making now: what to automate, what to keep with people, and what it costs to run.
How a programme is agreed
A group normally has 8–10 participants, so everyone builds, tests and takes part in the review.
- Tell us about the group, its current work and the capability you want to develop.
- Discuss technical starting points, suitable practice and the learning scope with Sunil.
- Agree live hours, independent work, group size, schedule and fee in a separate proposal.
For a product company this is usually one team at a time, tied to a feature or platform goal, with further groups only if the first one works.
Funding one person instead? Compare the individual Agentic AI programmes. An employer can sponsor a place.
Not sure it's training you need?
If one feature or architecture needs an outside review now, bounded advisory may fit better. For a company-wide view of where AI is worth investing, see the AI Readiness Assessment.
Other organisations: Global capability centres · IT services companies · All enterprise learning
Discuss your team
Tell us about the product, the team and what it needs to handle better.