The idea in one minute#
The role needs four kinds of skill: AI (what models can and cannot do), systems (how software and cloud fit together), business (value, cost, how organisations decide) and communication (writing, drawing, running a room). Nobody arrives with all four. Everybody arrives with one or two.
The route in is therefore the same for everyone and different for everyone: find your strong side, fill the weakest side to “good enough”, and prove it with artefacts — a design document, a cost model, a working prototype — not with a list of courses watched.
Architects are described as T-shaped: broad across all four, deep in one. This lesson gives you the map, a way to score yourself, and a study plan that fits around a job.
A picture#
flowchart TB
subgraph AI["AI"]
direction TB
A1[":i-brain: How models behave<br/><small>tokens, context, limits</small>"]
A2[":i-search: Patterns<br/><small>retrieval, agents, tuning</small>"]
A3[":i-flask-conical: Evaluation<br/><small>does it work, as a rate</small>"]
end
subgraph SYS["Systems"]
direction TB
S1[":i-cloud: Cloud basics<br/><small>compute, network, identity</small>"]
S2[":i-database: Data<br/><small>where it lives, who may see it</small>"]
S3[":i-shield-check: Security and reliability"]
end
subgraph BIZ["Business"]
direction TB
B1[":i-coins: Cost and value<br/><small>unit cost, ROI, TCO</small>"]
B2[":i-scale: Risk and regulation"]
B3[":i-building-2: How enterprises buy and decide"]
end
subgraph COM["Communication"]
direction TB
C1[":i-message-square: Discovery<br/><small>asking, listening</small>"]
C2[":i-file-text: Writing and diagrams"]
C3[":i-users: Running a room<br/><small>workshops, reviews</small>"]
end
AI --> T[":i-layers: <b>The architect</b><br/><small>broad in all four, deep in one</small>"]
SYS --> T
BIZ --> T
COM --> T
class A1,A2,A3 compute
class S1,S2,S3 memory
class B1,B2,B3 queue
class C1,C2,C3 io
class T neutralHow it really works#
The twelve skills, and what “good enough” means#
“Good enough” is the bar for the skills that are not your deep one. It is lower than people fear.
| Area | Skill | Good enough means you can… | Where this course covers it |
|---|---|---|---|
| AI | How models behave | Explain tokens, context windows and why answers vary, without mathematics | How AI Works, for Architects |
| AI | Patterns | Say when retrieval beats fine-tuning and when an agent is overkill | The Solution Toolbox |
| AI | Evaluation | Design a test set and a pass mark before the build | Evaluation as the Contract |
| Systems | Cloud basics | Read a cloud architecture diagram and ask about identity and network | The Cloud and Software Ground Floor |
| Systems | Data | Ask where the data is, who owns it, how fresh it is and who may see it | Discovery |
| Systems | Security and reliability | Name the trust boundaries and the failure modes of your own design | Security, Privacy and Residency |
| Business | Cost and value | Compute cost per task and payback period on a napkin | Cost, TCO and ROI |
| Business | Risk and regulation | Place a use case in the right legal risk class | Governance and Regulation |
| Business | How enterprises decide | Name the sponsor, the blocker and the buyer for any project | Discovery |
| Communication | Discovery | Run an hour-long interview and leave with numbers | Discovery |
| Communication | Writing and diagrams | Write a one-page decision record someone else can challenge | The Solution Design Document |
| Communication | Running a room | Get a group to rank ten ideas without the loudest voice winning | Qualifying and Prioritising Use Cases |
Score yourself#
Give yourself 0–3 on each of the twelve.
0 I could not explain it to a friend
1 I could explain it, but I have never done it
2 I have done it once, with help
3 I have done it more than once and could teach itAdd each area’s three scores (maximum 9). Your highest area is your spike. Any area under 3 is a gap that will embarrass you in a meeting and should be closed first.
Five common starting points#
| You are now | Your spike | Your likely gap | First three things to do |
|---|---|---|---|
| Software or backend engineer | Systems | Business, discovery | Learn unit economics; sit in on sales or customer calls; write a business case for something you already built |
| Cloud or infrastructure engineer | Systems | AI patterns, evaluation | Work through the Solution Toolbox; build one retrieval prototype; write its evaluation set |
| Data scientist or ML engineer | AI | Enterprise systems, cost | Learn identity and network basics; price a model you trained over three years, including people |
| Business analyst, consultant, product manager | Business, communication | AI behaviour, systems | Do the Foundations topic slowly; build one tiny thing end to end, however ugly |
| Student or career changer | None yet — which is fine | Everything, evenly | Follow the 24-week plan below in order; ship two artefacts |
A 24-week plan that fits around a job#
About six to eight hours a week. Each phase ends with something you can show.
| Weeks | Read | Do | Artefact you keep |
|---|---|---|---|
| 1–4 | The Role | Use three different AI assistants on the same real task; note where each fails | A one-page “what models are bad at” note in your own words |
| 5–8 | The Solution Toolbox | Build one retrieval prototype over your own documents with any tool | A working demo and a list of the ten questions it gets wrong |
| 9–13 | From Problem to Design | Interview one real person about a slow task at their work | Discovery notes, a scorecard, a cost model, a six-page design document |
| 14–18 | Production Architecture | Review your own design against the six pillars; write its evaluation gate | A review with findings, an evaluation plan, a risk register |
| 19–22 | Enterprise Strategy | Redo the design for 20 teams instead of one | A platform sketch and an operating model on two pages |
| 23–24 | The Career | Mock interview; choose one certification | A portfolio of four documents and one demo |
The “Do” column is not optional. Reading produces vocabulary; only doing produces judgement, and judgement is what the role is hired for.
The habits that separate architects from senior engineers#
- Write before you draw. A paragraph that states the problem exposes gaps a diagram hides.
- Ask “compared with what?” Every benefit is measured against today’s process, not against nothing.
- Carry three numbers. Volume per day, cost per task, and the accuracy the business can live with. If you do not know them, your next meeting is to find them.
- Prefer the reversible choice. When two options look equal, take the one that is cheaper to undo.
- Keep a decision log. One line per decision, with the date and the reason. In six months it is the only memory the project has.
What you do not need#
- A degree in machine learning. You need to predict how models behave, not derive them.
- To train a model from scratch. Useful once, for intuition; not part of the job.
- Every cloud. Learn one properly; the other two then take a week each.
- Every framework. Frameworks last a year. Patterns last a decade.
Remember this#
- Four areas: AI, systems, business, communication. Broad in all, deep in one.
- Close any area that scores under 3 out of 9 before deepening your spike.
- Artefacts prove skill: a design document, a cost model, an evaluation plan, a demo.
- Judgement comes from doing the exercise, not from reading the lesson.
Try it#
- Score yourself on the twelve skills. Write down your spike and your lowest area.
- Pick your row in the starting-points table and schedule the first of its three actions this week.
- Start a decision log today, in any tool. Enter your first decision: which cloud you will learn properly, and why.
Check yourself#
- What does T-shaped mean for an architect?
- Why does the plan insist on an artefact at the end of every phase?
- A strong ML engineer wants this role. Which area is most likely to be their gap, and what is the fastest way to close it?