Exercise 1: The Architect's Work

Overview

Four documents, all for Lintel, about two hours of work. They are the first entries in a portfolio you’ll build across all ten exercises, so write them as if someone else will read them — because in Exercise 10 you’ll assemble them into something that has to make sense as a whole.

There’s no instructor to grade these. The rubric at the bottom does that job, and it only works if you’re honest with it.

How to work

Do these in order. Each feeds the next, and the charter is genuinely easier last.

Timebox it — roughly 30 minutes each. These are first-pass artifacts, and Module 1 you is supposed to get some of it wrong. The stakeholder map in particular gets revisited in Exercise 8, and the gap between the two versions is part of the point.

Context and business-value brief — ~30 min
System context diagram — ~30 min
Stakeholder map v0 — ~20 min
Architecture charter — ~30 min
Self-check — five pass/fail items

Artifact 1 — Context and business-value brief

One to two pages. What this initiative is for, written so Dev Raghunathan and Maya Ortiz would both recognise their own concerns in it.

The test: an engineer should be able to read it and not find a single technology named.

Cover:

The problem, in business terms. Not “we need a RAG system” — what is happening, to whom, at what cost.
The drivers, with numbers. All four from the case, and be explicit about which the architecture serves first when they conflict.
The arithmetic. Work out what deflection is actually worth. Write down your assumptions — deflection rate, cost per interaction — because they’ll be wrong and you’ll want to know which ones.
Where the value really comes from. If the cost saving alone doesn’t cover the $1.2M, say so plainly and say what does.
At least one concrete consequence of unpredictable output. Not “the model might be wrong” — what specifically it would cost Lintel, and what the system does about it.
Success measures. How you’d know in six months whether this worked.

Artifact 2 — System context diagram

One diagram, C4 Level 1. The AI support platform as a single box, with everything it touches around it.

Include the people and systems that interact with it — and be thorough about the ones that are easy to forget. External model providers. The docs corpus. Zendesk. The Rails monolith. Grace’s 45 agents. The customers.

Do not draw anything inside the box. No retrieval service, no vector store, no orchestrator. If you catch yourself adding them, you’ve dropped to a different zoom level and lost what the diagram is for.

For each boundary crossing, note what data flows across it. Two of those crossings will make Lena Fischer’s job harder — find them now rather than in month six.

Artifact 3 — Stakeholder map v0

One page or one diagram. All seven people from the case, positioned by how much influence they have and how much they care.

Then, for each: what they say they want, and what you think they actually want. Sometimes those match. When they don’t, that gap is the most useful thing on the page.

Mark clearly who can stop this — and note that formal authority and actual power are different lists at Lintel.

Call this v0 and mean it. You’ve been here four weeks. You’re guessing, and Exercise 8 exists to correct it.

Artifact 4 — Architecture charter

One page. Not more.

Accountable for — the outcomes, not the tasks
I decide — be specific
I don’t decide — the section that carries the weight; at least four items
I advise on — where you have a voice but not a vote
How I work — how decisions get recorded, and how someone disagrees with you productively

Then answer this, in a sentence at the bottom: who needs to have read this for it to mean anything, and how will you get it in front of them? A charter nobody has seen is a diary entry.

Self-check

Apply each to your finished artifacts. Pass or fail — resist “sort of”.

Every architectural goal traces to a named business driver with a number. Not “improve efficiency” — which driver, and what figure.
The charter states what you will not decide, not only what you will. At least four items, specific enough to point at in an argument.
The context diagram shows external actors and systems, not internal components. If a reader can tell how the system works inside, it’s the wrong level.
The three architecture types are applied to Lintel specifically, not defined abstractly. Somewhere in your artifacts, real Lintel decisions are sorted into Solution, Enterprise and System.
At least one item names a concrete consequence of AI non-determinism for this system. Specific to Lintel, with a cost and a response.

If something failed

Fix it now rather than moving on. Exercise 2 builds directly on the business-value brief, and a vague brief produces a vague requirements catalog, which produces an architecture serving nothing in particular.

The most common failure is the first one — goals that sound like goals but have no number attached. Go back and put a figure on each. If you can’t find one, that’s a real finding: you’ve located something the business hasn’t decided, and asking about it is exactly the kind of question a new architect should be asking in month one.

→ The charter says what you own. Next, what the business actually needs from it. — Exercise 2, after reading Handbook Module 2.