Pidoku

The Skills Map and Your Route In

Foundations 40 min Difficulty 1/5 Lesson 02 of 04

Prerequisites What an AI Solution Architect Does

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 neutral

How 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.

AreaSkillGood enough means you can…Where this course covers it
AIHow models behaveExplain tokens, context windows and why answers vary, without mathematicsHow AI Works, for Architects
AIPatternsSay when retrieval beats fine-tuning and when an agent is overkillThe Solution Toolbox
AIEvaluationDesign a test set and a pass mark before the buildEvaluation as the Contract
SystemsCloud basicsRead a cloud architecture diagram and ask about identity and networkThe Cloud and Software Ground Floor
SystemsDataAsk where the data is, who owns it, how fresh it is and who may see itDiscovery
SystemsSecurity and reliabilityName the trust boundaries and the failure modes of your own designSecurity, Privacy and Residency
BusinessCost and valueCompute cost per task and payback period on a napkinCost, TCO and ROI
BusinessRisk and regulationPlace a use case in the right legal risk classGovernance and Regulation
BusinessHow enterprises decideName the sponsor, the blocker and the buyer for any projectDiscovery
CommunicationDiscoveryRun an hour-long interview and leave with numbersDiscovery
CommunicationWriting and diagramsWrite a one-page decision record someone else can challengeThe Solution Design Document
CommunicationRunning a roomGet a group to rank ten ideas without the loudest voice winningQualifying 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 it

Add 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 nowYour spikeYour likely gapFirst three things to do
Software or backend engineerSystemsBusiness, discoveryLearn unit economics; sit in on sales or customer calls; write a business case for something you already built
Cloud or infrastructure engineerSystemsAI patterns, evaluationWork through the Solution Toolbox; build one retrieval prototype; write its evaluation set
Data scientist or ML engineerAIEnterprise systems, costLearn identity and network basics; price a model you trained over three years, including people
Business analyst, consultant, product managerBusiness, communicationAI behaviour, systemsDo the Foundations topic slowly; build one tiny thing end to end, however ugly
Student or career changerNone yet — which is fineEverything, evenlyFollow 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.

WeeksReadDoArtefact you keep
1–4The RoleUse three different AI assistants on the same real task; note where each failsA one-page “what models are bad at” note in your own words
5–8The Solution ToolboxBuild one retrieval prototype over your own documents with any toolA working demo and a list of the ten questions it gets wrong
9–13From Problem to DesignInterview one real person about a slow task at their workDiscovery notes, a scorecard, a cost model, a six-page design document
14–18Production ArchitectureReview your own design against the six pillars; write its evaluation gateA review with findings, an evaluation plan, a risk register
19–22Enterprise StrategyRedo the design for 20 teams instead of oneA platform sketch and an operating model on two pages
23–24The CareerMock interview; choose one certificationA 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#

  1. Score yourself on the twelve skills. Write down your spike and your lowest area.
  2. Pick your row in the starting-points table and schedule the first of its three actions this week.
  3. Start a decision log today, in any tool. Enter your first decision: which cloud you will learn properly, and why.

Check yourself#

  1. What does T-shaped mean for an architect?
  2. Why does the plan insist on an artefact at the end of every phase?
  3. 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?

↑↓ navigate↵ openesc close

drag to pan · scroll to zoom