Skip to content
AI Engineer Guide.

concept · career · intermediate · concept 1 of 33

Role map

0 of 33 complete

Why it matters

Candidates apply to a title and prepare for a different job. An applied AI role that owns a product surface needs evaluation, latency and cost judgement; a platform role that serves models to other teams needs capacity, interfaces and incident response. The interview loop reflects the ownership, so reading the job description for what the team is accountable for is more useful than the title. Mapping roles also clarifies which gaps are worth closing.

Mental model

Roles are territories with different borders: the same word appears on the map but the responsibilities inside differ. The metaphor breaks because territories are described in job posts that are often aspirational, so you have to infer ownership from the team and the interviews.

How it works

Mechanism, step by stepInteractive
  1. 1Read the job description for the noun the team owns: a product, a platform, a dataset, or a research question.
  2. 2Identify who consumes their output, since internal consumers imply interface and reliability work.
  3. 3Map the interview loop to those responsibilities and note which round carries the most weight.
  4. 4Compare your evidence against that ownership, not against the title.
  5. 5Ask in the recruiter call what the team shipped in the last two quarters and what broke.
  6. 6Choose roles where your existing evidence matches the ownership, rather than where the title sounds closest.

Step 1 of 6

Read the job description for the noun the team owns: a product, a platform, a dataset, or a research question.

Step 1/6 · step 1

Figure. The mechanism in this page's own steps. The figure adds sequence, not new claims.

What to notice. A step can be read alone, but the order is the mechanism. The control lets you stop on any one.

Text version

The page's steps: 1. Read the job description for the noun the team owns: a product, a platform, a dataset, or… 2. Identify who consumes their output, since internal consumers imply interface and reliabil… 3. Map the interview loop to those responsibilities and note which round carries the most we… 4. Compare your evidence against that ownership, not against the title. 5. Ask in the recruiter call what the team shipped in the last two quarters and what broke. 6. Choose roles where your existing evidence matches the ownership, rather than where the ti…

  1. Read the job description for the noun the team owns: a product, a platform, a dataset, or a research question.
  2. Identify who consumes their output, since internal consumers imply interface and reliability work.
  3. Map the interview loop to those responsibilities and note which round carries the most weight.
  4. Compare your evidence against that ownership, not against the title.
  5. Ask in the recruiter call what the team shipped in the last two quarters and what broke.
  6. Choose roles where your existing evidence matches the ownership, rather than where the title sounds closest.

Trade-offs

ChoiceGood forCostFails when
Product AI roleVisible impact and fast feedbackOn-call for quality and costYou prefer deep platform work
Platform roleLeverage across teamsInterface negotiation and less product contactYou want direct user feedback
Data and retrieval roleDeep correctness workPerceived distance from the modelYou want modelling and training
Research roleNovel problemsSlow feedback and publication pressureYou want shipped software
Trade-off mapInteractive

Good for

Visible impact and fast feedback

Cost

On-call for quality and cost

Fails when

You prefer deep platform work

  1. You prefer deep platform work
  2. You want direct user feedback
  3. You want modelling and training
  4. You want shipped software

Figure. Each choice with what it buys, what it costs, and the case where it loses.

What to notice. The fourth column is the decision: if the losing case describes your situation, the choice is wrong.

Text version

Four trade-offs: Product AI role is good for Visible impact and fast feedback, costs On-call for quality and cost, and fails when You prefer deep platform work. Platform role is good for Leverage across teams, costs Interface negotiation and less product contact, and fails when You want direct user feedback. Data and retrieval role is good for Deep correctness work, costs Perceived distance from the model, and fails when You want modelling and training. Research role is good for Novel problems, costs Slow feedback and publication pressure, and fails when You want shipped software

Failure modes

  • Title matching. Preparation targets a title while the loop tests another ownership. Debug: list the responsibilities in the job post and check which interview round covers each.
  • Ignoring consumers. A platform role is treated as a product role and the reliability questions go unanswered. Debug: ask who depends on the team's output and what happens when it fails.
  • Aspirational job post. The description promises research and the team does maintenance. Debug: ask what the last two quarters actually shipped.

In production

Keep a shortlist with the ownership, the loop and your evidence gap per role. Revisit it after each interview, since the debrief reveals what was actually weighted.

Common misconceptions

People think: All AI engineer roles are the same job with different titles.

Actually: Ownership differs enough to change the interview loop and the day-to-day work.

People think: A research title is always more senior.

Actually: Scope and impact vary; a platform role can carry more organisational leverage than a narrow research one.

Interview angle

Strong answers connect the role's ownership to the evidence they have and name the gap honestly. Weak answers treat the title as the job and prepare one generic story for every loop.

Sources