About

About Sunil Mathew

Sunil Mathew teaches the design of agentic systems and advises engineering teams making those decisions. He brings engineering and leadership experience from Google, Amazon, Walmart and startups. His focus is the reasoning behind a system’s design and the evidence that helps a team make its next decision.

This page states what can be sourced. Where a fact is outstanding it is named rather than filled in.

Who this page is for

You are deciding whether this transfers

Two readers arrive here. One is weighing thirty live hours against everything else that could fill them. The other has a design decision in front of a team and is deciding whether an outside reviewer would help or merely add an opinion.

Both are asking the same question, and it is not about credentials. It is whether the way this person reasons about a system transfers to the system you are responsible for. That question is answerable, so the rest of this page answers it rather than listing achievements.

An illustrative example

How a design question gets answered

A returns assistant reads a policy, checks a request and recommends a refund. Someone asks whether it can issue the refund as well. On a diagram that is one more connection. In production the system can now move money.

The example is illustrative and no customer incident is claimed. What follows is the order the questions get asked in — the same order the programme and an advisory review both use.

  1. 01

    Separate the recommendation from the action

    Gathering policy evidence and preparing an answer is one responsibility. Moving money is another. They look like one box and they need different controls.

  2. 02

    Ask what evidence permits the action

    Not what the prompt says is permitted. What the tool itself will refuse to do when the evidence it was handed does not hold up.

  3. 03

    Decide what happens when the answer is uncertain

    A request outside the policy needs a clear next step. A tool that times out needs a way to examine the state of the action before repeating it.

  4. 04

    Ask what you can see afterwards

    Which evidence was used, what the tool did, whether the workflow stayed inside its boundaries. Without that, evaluation cannot tell you which decision to change.

Each of those four is worked through in full in the published guides, which is the most direct evidence available of how the reasoning actually runs.

What is verified

The claims on this site, and their source

Experience Engineering and leadership at Google, Amazon, Walmart and startups
Based Bengaluru, India
Focus The reasoning behind a system’s design, and the evidence behind a team’s next decision
Teaching 6 weeks, live, at ~5 hrs / week. Live, with a working build
Elsewhere LinkedIn

Next step

Two ways to work together

A specific question that fits neither? Contact lists which route goes where.

Written . Pending Sunil’s factual approval.