Edition 1 · Q4 2026 · a quarterly series

The roles teams are learning to interview for

Companies from OpenAI to Stripe now hire AI engineers, forward deployed engineers, GTM engineers and SEO managers who answer for AI search, with no settled way to interview them. The postings rarely say how: none of about 60 forward deployed engineer postings we read in October 2026 did. This edition sets out what each loop has to test.

Published . Postings read on employers’ own job boards in October 2026, and every source rechecked . Written by Hirezen, which publishes the interview plans this page links — how it was made.

The four roles at a glance

RoleWhat the title coversWhat the loop has to testWhat employers publish about the interview
AI engineerThree jobs: product features, internal tools, customer-facing work at model companiesHow often the feature is wrong, what fails and what is reused, what it may do with hostile inputA split on AI in the interview: Anthropic says no unless told, Canva expects it
Forward deployed engineerEngagements owned end to end, or engineers placed by staffing firmsDiscovery with a customer who volunteers nothing, and what ran after a past engagementNone of about 60 postings described the interview; Column publishes its process
GTM engineerAt least four jobs across 31 employers, from sales agents to SalesforceRouting a signal to a seller, an outbound month, a CRM sync in code3 of 38 postings described a process; Clay publishes its take-home
SEO manager for search and AI answersSearch and AI answers, measured as citations rather than rankA live answer engine, a page's raw HTML, an executive cutting the budgetJuniper Square asks for one citation the candidate earned, and how

What the four loops have in common

  1. Employers rarely write the loop down

    On October 6, 2026 we read about 60 forward deployed engineer postings on employers' own job boards, and none described the interview process. Of 38 GTM engineer postings read in full on October 8, three described a process, all for customer-facing variants of the role. The few employers who publish a loop do it on purpose. Column's hiring page sets out a 30-minute recruiter screen, typically two or three 30–45 minute interviews, for most roles a take-home case study of typically two to four hours, and a final round presenting it to a panel, and says the process can vary by role (Column). Clay publishes the take-home it uses for GTM engineers (Clay). A team hiring one of these roles for the first time has little to copy.

  2. One title covers several jobs, so pick one before writing the loop

    AI engineer means at least three jobs, and when we read GTM engineer postings on October 8, 2026, the title turned up at 31 employers for at least four, from software engineers building sales agents to Salesforce engineering. Both are set out below. The forward deployed title is used by companies whose engineers own each engagement and by staffing firms that place engineers into them. A loop written for every meaning reads none of them well, so each loop linked here says which job it is for, and what to add for the others.

  3. The vocabulary arrived before the practice, so hand the candidate an artifact

    Almost every question about these roles has a fluent answer in a blog post, so a round built on vocabulary measures reading. The rounds linked here each hand the candidate something to work on and watch what they notice in it: a pack of model traces, a page's raw HTML that disagrees with what the browser shows, a sample coverage report claiming 47 placements of which 31 are copies of one wire release, an ad platform and a warehouse reporting 1,200 and 780 conversions for the same month.

  4. The moments that decide the job happen out of sight, so recreate one

    A forward deployed engineer's first week of discovery, or an SEO manager's meeting with the executive who sets the budget, never happens in front of an interviewer. Two rounds here recreate one live. In one, an interviewer plays a customer who volunteers nothing, and the candidate has to find out what matters. In the other, the executive has already decided to cut the budget, and the decision goes against the candidate partway through. These cost more to run, the customer simulation needing two interviewers, and they are the rounds most likely to separate a candidate who has done the work from one who has watched it.

  5. Decide the AI rule before the first round

    Employers disagree about whether a candidate may use AI in the interview. Anthropic asks candidates to complete take-homes and live interviews without AI unless told otherwise (Anthropic, updated July 2025). Canva wrote in June 2025 that it expects backend, machine learning and frontend candidates to use AI tools in its technical interviews, and looks at how they use them, break down ambiguous requirements and fix flaws in AI-generated code (Canva). Either is defensible. Leaving it unsaid is not: in a round that credits what a candidate names unprompted, an assistant can name it for them.

Role by role

Role 1 of 4

AI engineer

An AI engineer builds products on foundation models rather than training them. The work is a feature that is wrong some of the time by construction, so the interview has to find out whether the candidate knows how often, makes the failure cheap, and keeps the product standing when the model is down or has been talked into something.

LinkedIn ranked AI engineer first among the fastest-growing US job titles in its Jobs on the Rise list, published in January 2026 from jobs members started between January 2023 and July 2025 (CNBC). The title covers three jobs. At Brex it means building the workbench where finance teams work with its agents, and owning those features through production (Brex). At Stripe it means building tools for the solutions architects (Stripe). At Anthropic an applied AI engineer helps enterprises deploy its models into their operations and products (Anthropic), which is closer to a forward deployed engineer. Training and fine-tuning stay with the machine learning engineer, the line Latent Space drew in The Rise of the AI Engineer in June 2023. The loop is written for the first job.

What the loop hands the candidate, round by round

  1. 1. Coding: degrade, not fail

    The code around one model call, and what it does when the call fails.

  2. 2. System design: caching and reuse

    What may be reused across users, and what a wrong reuse costs.

  3. 3. Evals and measurement

    Whether any of the numbers reported about the feature are true.

  4. 4. System design: untrusted input and tool permissions

    What the system may do when the text it read was written by someone hostile.

Role 2 of 4

Forward deployed engineer

A forward deployed engineer fits their employer's product into one customer's systems, data and people, and leaves behind something the customer can run. Candidates arrive from solutions engineering, consulting and contracting with fluent stories about engagements, so the interview has to find out what ran afterward and who decided what.

In May 2026 OpenAI launched the OpenAI Deployment Company to embed forward deployed engineers in customer organizations, and agreed to acquire Tomoro, an applied AI consulting firm whose acquisition, subject to closing, would bring about 150 forward deployed engineers and deployment specialists (OpenAI). Elsewhere the title is splitting by domain and language: Harvey hires a forward deployed finance engineer for private credit (Harvey), and Decagon German-speaking agent deployment engineers (Decagon). Column, a bank hiring its first forward deployed engineer as the technical owner for strategic customers from onboarding to launch (Column), is one of the few employers to publish how it interviews.

What the loop hands the candidate, round by round

  1. 1. Customer scoping simulation

    A new engagement, live, with an interviewer playing a customer who volunteers nothing.

  2. 2. Engagement deep dive

    A real engagement the candidate chose, reconstructed: what ran afterward, and who decided.

  3. 3. Coding: the customer export

    An export from a system the candidate cannot change, with a security review waiting behind it.

  4. 4. System design: untrusted input and tool permissions

    What an agent may do on someone's behalf, and the morning it does too much.

Role 3 of 4

GTM engineer

A GTM engineer builds the systems that turn a buying signal into a seller's next action: matching, routing, enrichment, automated outbound and the CRM. Every candidate can demo a workflow that looks good, so the interview has to find out who used it, what it moved, and what it did to the CRM afterward.

The 31 employers posting the title in October 2026 meant at least four jobs by it. OpenAI's GTM Growth Engineering team builds products tied to pipeline quality and to sellers' and marketers' productivity (OpenAI). Sierra's GTM engineer owns the enriched, scored record that lands in Salesforce, and ships production code through review (Sierra). Ramp's GTM business systems engineers build its Salesforce quoting platform and need seven or more years on Salesforce (Ramp). Artisan's forward deployed GTM architects plan customer rollouts, and their interview includes a 60-minute working session designing one for a real enterprise scenario (Artisan). Clay, which sells a data enrichment and workflow tool many GTM engineers use, says it coined the title in 2023 (Clay), and recommends a behavioral interview, several interviews on how a candidate handles ambiguity, and a take-home (Clay). The loop is written for Sierra's kind of job: the engineer who owns the path from a signal to a seller.

What the loop hands the candidate, round by round

  1. 1. System design: signal to rep

    Where an inbound request should go, from a brief that leaves out who owns what.

  2. 2. Work sample: an outbound workflow's first month

    Someone else's outbound workflow, and its first month of numbers.

  3. 3. Coding: the signup sync

    Signups into a CRM in code, under a rate limit other tools share, over fields reps have edited.

  4. 4. Behavioral: one system

    A system the candidate built, and what the sellers did with it.

Role 4 of 4

SEO manager for search and AI answers

An SEO manager now owns how a company is found in Google and in the answers ChatGPT, Perplexity, Gemini and Claude give. Any candidate can talk about entities and share of voice, so the interview has to find out whether their measurement held up when an answer engine gave different answers to the same question.

Postings now name the answer engines beside Google, as AEO or GEO. Stripe's head of AEO and SEO owns visibility in general-purpose assistants and in coding assistants such as Claude and Codex (Stripe). Juniper Square asks its SEO and AEO manager to point to a specific citation they earned in an AI answer engine and explain exactly how they did it (Juniper Square), the most concrete evidence any posting we read asked for. The loop is written for a company that already gets a real share of its demand from search, and an SEO manager who works with web engineers and a content team.

What the loop hands the candidate, round by round

  1. 1. What the answer engine said

    A live answer engine, then a predecessor's dashboard, and a metric precise enough to hand to an analyst.

  2. 2. The entity and the citation

    One AI answer that cites four sources and not the company, and a fixed quarter to spend.

  3. 3. What the crawler received

    One URL whose raw HTML disagrees with what Chrome shows, and a test that stayed green through it.

  4. 4. The budget defense

    A role-play with the executive who has already decided to cut the budget.

Four more rounds for the team around the SEO manager

Written in August 2026 for the seats that work beside an SEO manager. This edition has not read employers’ postings for these titles, so it says only what each round hands the candidate.

  • A site migration someone else scoped: a redirect map generated from a crawl, a validation job that passed, and organic traffic down after cutover.

  • A sample library of about 1,400 pages, about 60 of which bring most of the visits, and what to stop publishing.

  • A sample coverage report claiming 47 placements, most of them syndicated copies, with organic and referral traffic flat.

  • Marketing analytics engineer

    The number two systems disagree about

    An ad platform reporting 1,200 conversions and a warehouse reporting 780 for the same month, when neither is broken.

How this edition was made

This edition reads postings on employers' own job boards, through the public feeds Greenhouse, Ashby and Lever publish, and the pages employers publish about how they hire. Its counts are of postings we read, on the dates given: about 60 forward deployed engineer postings on October 6, 2026 and 38 GTM engineer postings on October 8. SEO postings were read on October 23. Every outside page cited here was opened again on November 16, 2026, and postings that had closed since October were swapped for live ones or left out.

It quotes no job-board search totals. They move too much from week to week to report as a trend, so this edition counts only postings it read, and cites growth only where the publisher of the number stands behind it.

The interview plans it links were written in August 2026 and are free to read. Hirezen makes interview software and wrote this edition; the plans are Hirezen's own.

A correction gets a dated note at the foot of this page. The next edition, for the first quarter of 2027, carries anything that has changed.

Every plan linked here is free to read, and free to run in Hirezen for one open role at a time: the questions, what each one is for, and the signals your interviewers mark as they hear them.