Agentic Code Engineering - ACE

Sachin Mundra

Agentic Code Engineering, SmartGenie

Sash Mohapatra

Ex-Microsoft Data & AI | 2,000+ trained

Agents that pass the demo fail in production. ACE fixes the discipline.

Almost every agentic course teaches you to make an agent do something impressive. Very few teach you to make one you can trust.

AI coding tools are confident, fast and often wrong. A green result from an agent tells you it says it worked, not that it did. The gap is not intelligence, it is discipline: the contracts, reviews and evidence that software engineering learned the hard way, and the current wave skipped.

ACE revives that discipline for agentic systems. Over four weeks you build one real system on your own project: designed before it is generated, contract-first, tested with evals you write before the build, and put through an independent gate before it ships. Every claim it makes is backed by evidence a reviewer, or an auditor, could follow.

You leave with a working, gated agentic system, the templates to repeat it, and a method your team can adopt. It is the difference between a demo and something you would let near production.

What you’ll learn

Go from AI demos to a gated, working agentic system you can reason about, test and defend, backed by evidence, not vibes.

Use Anthropic's agent patterns (routing, orchestrator-workers, evaluator-optimiser) as a shared design vocabulary.
Design the agent loop, tools and tool contracts so behaviour is bounded and inspectable.
Budget the context window and keep long-term project memory the way a disciplined team does.

Write a mature architecture document and an honest assumptions register before building.
Model contracts, seams and shared types first, so integration is designed, not discovered.
Specify behaviour as a contract before implementation, and refuse to fabricate a missing metric.

Reframe the testing pyramid as an evaluation pyramid and write the check before the build.
Run a verification ladder so a passing result is backed by real command output, not a ticked box.
Build an evaluation harness for a real failure mode of your own agent, layered like the testing pyramid.

Subject your design to an independent gate before you build, and treat a refusal as the product working.
Produce an evidence trail a reviewer, or an auditor, could actually follow.
Design and calibrate your own review rubric, then gate a peer's architecture and defend your verdict.

Build one real agentic system across the course: contract-first, independently gated, fully evidenced.
Present it for peer gate review to earn the ACE Practitioner certificate.
Take away starter templates: architecture skeleton, assumptions register, verification ladder and eval harness.

Learn directly from Sachin & Sash

Sachin Mundra

Contact

Production-ready AI · Governed, Secure, Auditable · Ex-Microsoft 20+ yrs.

Microsoft

Sash Mohapatra

Contact

20 years at Microsoft, engineer to Director of Data & AI Sales, Capital Markets

See all products from Sachin

Who this course is for

Software engineers moving into agentic development who want to do it properly, not by trial and error.
Engineers in regulated or high-stakes environments who cannot ship on vibes and need evidence and structure.
Technical leads who want a shared, disciplined method their team can actually follow.

What's included

Live sessions

Learn directly from Sachin Mundra & Sash Mohapatra in a real-time, interactive format.

Lifetime access

Go back to course content and recordings whenever you need to.

Community of peers

Stay accountable and share insights with like-minded professionals.

Certificate of completion

Share your new skills with your employer or on LinkedIn.

Maven Guarantee

Your purchase is backed by the Maven Guarantee.

Course syllabus

Week 1

Dec 1—Dec 6

Session 1. Foundations: from demo to disciplined agent

6 items

Week 2

Dec 7—Dec 13

Session 2. Design before code

5 items

Session 3. Contracts first and vertical slices

4 items

Schedule

Live sessions

2-4 hrs / week

Will need time commitment for the learning time and labs

Projects

2 hrs / week

Each session ends with a lab on your own project: stand up your agent loop, write the architecture document, build a vertical slice, an evaluation harness, and finally gate a peer's architecture. Together they build the one gated agentic system you present as your capstone.

Async content

1 hr / week

Optional homework with worked solutions, plus starter templates for the architecture skeleton, assumptions register, verification ladder and evaluation harness.