Technical and adjacent routes

What you will be able to do

See that AI teams need more than engineers and scientists, and how much coding each route really needs. Some of these routes are the fastest way into AI for people with experience in another field.

Before you startWhat to know and to have readyAbout 25 minutes

What you need to know

• The six job families and what each one produces, from The main job families: Six job families.

What you need

• Your notes.md file in the ai-career folder. You made it in AI in everyday products: Set up your course notes. If it is missing, make it again as in that section. This lesson adds the heading ## Coding levels.

The roles around the core

An AI feature can work perfectly and still fail. Nobody may need it, nobody may trust it, it may break a law, or nobody may notice when it gets worse. The roles in this lesson prevent these failures. We call them adjacent routes: close to the technical work, with less coding, but not less important.

Data analyst

Data analysts turn data into numbers and charts that people use to make decisions.

• Produces: reports, dashboards and answers to business questions, for example “How many support questions are about returns, and is that growing?”
• Coding: SQL and spreadsheets every day. Some analysts use Python.
• Why it matters for AI: analysts measure whether an AI product changed anything for the business. Many data scientists started as analysts.

AI product manager

An AI product manager decides what to build and why, and makes sure that it helps users.

• Produces: the problem statement, priorities, the success measures, and decisions such as “We start a pilot with 10 agents in week 6.”
• Coding: usually none. But they must understand how AI fails: models are sometimes wrong, cost money for each use, and change when the data changes.
• Why it matters for AI: most failed AI projects solved the wrong problem, or measured the wrong thing. Choosing well is the product manager’s job.

Evaluation specialist

An evaluation specialist checks how well an AI system works, before and after release. Evaluation means measuring an AI system’s outputs against an agreed standard.

• Produces: test sets of realistic inputs with correct answers, scoring rules, and reports such as “88 of 100 answers correct. 5 answers promised a refund that the policy does not allow.”
• Coding: a little to medium. Some evaluation work is careful reading and judging. Larger test sets need Python.
• Why it matters for AI: language models write fluent text even when the text is wrong. Without careful evaluation, nobody knows how often that happens.

AI governance and risk

Governance and risk specialists make sure that AI is used legally, safely and fairly.

• Produces: policies for AI use, risk assessments, checks before release, and records that show what the organisation decided and why.
• Coding: none to a little. Knowledge of law, privacy, risk or audit is more important.
• Why it matters for AI: new laws and customer expectations require organisations to show how their AI makes decisions and who is responsible.

Subject-matter expert

A subject-matter expert (SME) is a person with deep knowledge of the field where the AI is used. Examples are a claims handler in insurance, a nurse in a hospital, or a support lead at a help desk. At Larkfield, the SME is the support lead, who knows every policy and every exception.

• Produces: the definition of a correct answer, test cases, reviews of the AI’s outputs, and the rules that the team did not know.
• Coding: none needed for the role. Some SMEs learn technical skills and move into the roles above.
• Why it matters for AI: this role is so important that Module 2 has two lessons about it.

How much code does each route need?

Use this scale to compare routes. The levels describe what a person does with code in a normal week.

Level	What it means	Typical routes
0 — None	Uses AI tools and documents, but writes no code.	Subject-matter expert, many governance roles
1 — Reads and configures	Reads simple code and settings, uses no-code tools and spreadsheets with formulas.	AI product manager, some evaluation roles
2 — Writes scripts	Writes SQL queries and short Python scripts for analysis.	Data analyst, evaluation specialist
3 — Builds software	Writes code that other people depend on, with tests, version control and reviews.	Data engineer, ML engineer, AI engineer
4 — Research code	Writes experimental code for new methods, with advanced maths.	Researcher

Data scientists are usually at level 2 or 3. Solution architects read at level 3 and write at level 2.

The levels are not a ranking of value. A product manager at level 1 can make the decision that saves a project. A level-3 engineer who builds the wrong thing well wastes months. The levels tell you how much coding practice a route needs, so you can plan your time.

Guided practice: what level does the task need?

Here is a worked example. The task: “Every Monday, count last week’s support questions by topic, and show the trend in a chart.”

This needs a query on the help desk data and a chart. One short SQL query or a spreadsheet does it. Nobody depends on the code as a system. That is level 2.

Now do these yourself. Write the level for each task in notes.md under ## Coding levels.

1. Read 30 answers from the document assistant, and mark each one as correct, partly correct or wrong, with a reason.
2. Make the assistant’s panel show the date of the policy next to every answer.
3. Decide whether the assistant may answer questions about payments, or only about returns and delivery.
4. Write a script that sends 500 test questions to the assistant and counts the correct answers.

Show the answers

1. Level 0. This is careful judgement by a person who knows the policies, often an SME or an evaluation specialist. No code.
2. Level 3. It changes the product that the agents depend on, so it needs tested, reviewed code.
3. Level 0 or 1. It is a product decision. The product manager makes it, with the support lead and the governance specialist.
4. Level 2. It is a script for measurement. Nobody else depends on it as a product.

Independent variation: write the level where you are today, and the level of the route that interests you most. The difference is part of your learning plan in Module 4.

Common mistake

Thinking that an adjacent role needs no AI knowledge

Symptom: a product manager promises the director that the assistant “will always give the correct answer”. Two weeks later, an agent finds a wrong answer, and the director loses trust in the project.

Cause: the product manager knows the product, but not how AI behaves. Language models are sometimes wrong, even when they sound sure.

Fix: adjacent roles need little code, but they need AI literacy: how models learn, how they fail, and how teams measure them. Promise a measured result instead, for example: “In our test, 95 of 100 answers were correct. Every answer shows its source, so the agent can check it.”

Check your understanding

Which route would you suggest to a person who has worked as a support lead for ten years, does not code, and wants to work on AI projects soon? Why?

Show the answer

Good first routes are subject-matter expert on an AI project in customer support, then evaluation specialist or AI product manager for support tools. These routes use their deep knowledge of support work every day and need little or no code (levels 0–1). They still need AI literacy. Later, they can learn to write scripts (level 2) if they want a more technical role.

Check your understanding

An evaluation specialist reports 'the assistant is 92% accurate'. What question should the product manager ask next?

Show the answer

“What are the 8% of wrong answers, and how bad is each one?” A wrong answer about a delivery time is a small problem. A wrong promise of a refund costs money and trust. The same accuracy can be acceptable or not, depending on which mistakes it contains.

Recap

• Adjacent routes are close to the technical work, with less coding: data analyst, AI product manager, evaluation specialist, AI governance and risk, and subject-matter expert.
• Coding needs go from level 0 (none) to level 4 (research code). The level is not the value of a role.
• Every route needs AI literacy: how AI learns, fails and is measured.

Next, you see all these people work together on Larkfield’s assistant, including the week when something goes wrong.

Next: One project, many roles