Emerging Roles, and Reading a Title

Overview

A crop of new titles has appeared, and they are a mixture of genuine specialisations, sensible renamings, and job advertisements written by someone who has not thought about the work.

Telling them apart matters for a practical reason: a title tells you very little, and the cost of discovering that after accepting a role is high.

The titles

AI Architect / AI Solution Architect. Mostly a solution architect whose systems include probabilistic components. The genuinely new content is real but narrower than the title implies — evaluation discipline, inference cost modelling, non-determinism in requirements, and the safety and blast-radius questions that follow from systems taking actions. Modules 4 through 7 of this guide are that delta.

Whether it is a distinct role or a specialisation depends entirely on the organisation. At a company where this is the only such initiative, it is a solution architect with extra skills. At one building many such systems, it is a specialisation with its own practice.

Prompt Architect. The most questionable of the three as a standing role. Prompt design is a genuine skill and a small one, and it is unstable — a large part of what was prompt engineering two years ago is now handled by better models and better tooling. A role defined by it has a narrow and shrinking basis.

Where the title sometimes means something more durable: an organisation using it for the person who owns evaluation, context design and the standards that govern how the whole team’s systems are prompted. That is real work — and it is closer to what the next two topics describe than to writing prompts.

Chief AI Officer. Genuinely new in the narrow sense that no previous title carried it, and genuinely redundant in most organisations that create it. The function — deciding what the company does with a new capability class, and being accountable for it — is executive work that a CTO can hold. A separate officer makes sense where the regulatory and risk surface is large enough to be a full-time accountability, and otherwise it tends to be a signal to a market rather than a response to a need.

And the unglamorous one with real demand: integration architects. Most AI value in established companies comes from connecting a capability to systems and processes that already exist, under constraints nobody chose. That is Section 1’s integration setting, and the demand for it is larger and steadier than the demand for the newer titles — it is simply less interesting to write about.

graph TD
  A["<b>AI Solution Architect</b><br/>solution architect +<br/>evals, cost, safety"] --> D["How much is<br/>genuinely new?"]
  B["<b>Prompt Architect</b><br/>narrow and shrinking —<br/><i>unless it means<br/>evals and standards</i>"] --> D
  C["<b>Chief AI Officer</b><br/>new title, existing<br/>accountability"] --> D

  classDef real fill:#7D9A4E,stroke:#5F7A38,stroke-width:2px,color:#ffffff
  classDef thin fill:#E8A33D,stroke:#C4832A,stroke-width:2px,color:#ffffff
  classDef hub fill:#5B8FB9,stroke:#4A7A9E,stroke-width:2px,color:#ffffff
  class A real
  class B,C thin
  class D hub

Three titles, and the delta over an existing role is smaller in each case than the name suggests.

Reading a title against its content

Five questions that establish what a role actually is. They work in an interview, and they work on your own current position.

What decisions does this role make, and which does it only advise on? The single most informative question, and the one that distinguishes an architect with a remit from a consultant with a title.

Who can overrule it, and on what grounds? Module 8’s authority mapping. “Nobody” is a warning sign rather than a good answer — it usually means the role has no real stake in anything.

What does the role own that persists? A practice, a set of standards, an evaluation instrument, a documented architecture. A role owning nothing durable is an advisory position regardless of its name.

What would failure look like in two years? If nobody can answer, the role has no success criteria, which means its evaluation will be political.

And: is there an existing initiative, or is this the first? Being the first is a very different job — Module 8’s no-mandate condition — and it should be chosen deliberately rather than discovered.

What this means for how you describe yourself

The symmetric point, and it matters for the capstone’s executive summary.

Describing yourself by titles held communicates little, for exactly the reasons above. Describing yourself by the class of decisions you have owned and the instruments you have left behind communicates the thing a hiring manager is trying to establish — and it is also what the previous topic identified as the real measure of trajectory.

What the role here would be advertised as

A reading of this guide’s own role against the five questions.

The likely advertisement. “AI Solution Architect — lead the design of our AI-powered support platform.” Plausible, and it tells a candidate almost nothing about the work.

What the five questions reveal.

Which decisions does it make? Technical structure, the action boundary, the evaluation gate. It advises on scope, timeline and capacity, and owns none of them.

Who can overrule it? Lena Fischer on security and privacy, with a formal block. Ivan Petrov on anything touching the monolith, with no formal veto and an effective one. Dev Raghunathan on the business case. Tomas Beck on capacity. Four people, three of them able to stop the initiative, and that is the honest answer Module 8 produced.

What does it own that persists? After nine modules: an evaluation set, a tiered action list, a documented decision history, a documentation practice and a cost model. That is a substantial durable holding, and none of it was in the job description.

What would failure look like in two years? The platform exists, deflection is below target, and nobody can say whether the corpus or the technique is responsible — which is precisely the failure the evaluation work was built to prevent, and the 2023 episode was its rehearsal.

Is this the first? Yes, and the last attempt was switched off quietly. The no-mandate condition in its hardest form.

The delta over a plain solution architect role. Four modules’ worth: evaluation discipline, inference cost modelling, non-deterministic requirements, and the safety and blast-radius work that follows from a system taking actions. Real, and roughly a third of the job. The other two thirds were requirements, stakeholders, organisational reading, justification, process and evolution — which is a solution architect’s work and has not changed.

So the honest title is “solution architect, with the AI delta” — and the useful thing to say in an interview is not the title but the five answers above.

Feeds the workbook

Use this to frame the executive summary: the class of decisions owned and the instruments produced, rather than the titles of the exercises.

→ The role is clearer. Some of your colleagues in it will not be people. — Working with AI teammates