Pick the seat: a real target, not a vibe

The market for delivery managers entering AI is real. The market for delivery managers pretending to be staff engineers is also real, and it is a meat grinder. You pick a seat by looking at the last three years of your calendar, not at a compensation spreadsheet. Loved client-site mess, making a slice work, being the person in the room with a half-broken process and a deadline? That energy maps to Forward Deployed Engineer and to Solutions. Loved steering, RAID, vendors, and saying no to a portfolio of slogans? That energy maps to AI Delivery Lead and to Transformation. Loved speccing with engineers, cutting scope, and writing the document that made a release possible? That energy maps to AI Product Manager. There is no wrong answer. There is a wrong interview.

Write the last three years down as evidence, not as adjectives. "I am good with stakeholders" is not evidence. "I ran a 14-month claims-ops program with four vendors, a hostile operations director, and a go-live that slipped twice; I owned the RAID, the steering pack, and the cut-scope decision in month nine" is evidence. The first sentence could be anyone. The second sentence tells a hiring manager which rooms you have already survived. Do this for each of the five seats: two bullets of actual work that would look like that seat if you relabeled them.

Energy is not the same as competence. You may be competent at all five. You will not have energy for all five in a six-week loop. FDE loops ask you to be curious about a repo at 9pm. Solutions loops ask you to be curious about a discovery call after three others that day. Delivery loops ask you to be curious about a RAID log that nobody else wants to update. Transformation loops ask you to be curious about a kill list that will make a VP angry. AI PM loops ask you to be curious about a spec that an engineer will tear apart. Pick the curiosity you already have, not the one you wish you had.

If you will not write a 40-line script or sit in a repo, do not pick FDE as primary. That is not a moral failing. It is a scheduling fact. FDE hiring managers will put you in a technical screen or a take-home that assumes you can open a file, change a constant, and talk through a retrieval loop. If that prospect fills you with dread after week 2, believe the dread. If week 2 felt like a relief — "finally, a thing I can make" — then FDE is in play, even if your GitHub is empty. Empty is allowed. Dread is not, because it will show.

If you wilt in pre-sales theatre, do not pick Solutions as primary. Solutions is discovery, objection handling, a whiteboard, and a champion who needs a demo that looks like their world. You already do some of this in delivery. The delta is that the room is a deal, not a program, and the clock is a quarter, not a year. Some delivery managers love that clock. Some find it oily. Be unromantic. A solutions interview will ask you to qualify, not to pitch. If you cannot imagine saying "this use case is a no, here is a thinner one that might be a yes," pick another seat.

AI PM is not "I like products." It is "I have sat with engineering and cut scope, and I can write a contract for a feature that includes evals and a refusal policy." If your last three years were status reporting with no design authority, AI PM will feel like a costume. Transformation is not "I think strategically." It is "I have killed work, scored a portfolio, and survived the politics of no." If your last three years were a single workstream with no portfolio, Transformation is a stretch for now — a backup at most, a later move after you have run an AI delivery seat.

If you cannot decide after the matrix, the default is Primary: AI Delivery Lead. Backup: Solutions Engineer. That pair uses your existing proof — programs, RAID, vendors, steering, discovery workshops — and this month's literacy — RAG, evals, HITL, cost — without requiring you to fake a CS degree or a product-org pedigree. It is the closest leap, and closest leaps get offers. Exotic leaps get "we went with someone who has done the job."

Move FDE to primary only if week 2 felt like a relief and you will, in the next four weeks, open a repo or a notebook and make the wiki copilot retrieve on a folder of real documents. The hiring bar at many FDE shops is not LeetCode. It is "can this person sit at the customer, design a slice, and not wait on a back-office engineer for every loop." Delivery managers who can call an API, sketch RAG, and write an eval are in a thin talent pool. That is the bet. It is only a bet if you will actually touch the tools.

Keep Solutions as backup even if FDE is primary. Many FDE second rounds are customer-steering simulations. Many solutions first rounds look like FDE first rounds with a sales partner in the room. The artifacts you build this week — demo story, operating pack, discovery no — serve both. Transformation as a backup is sensible if you have already run a PMO or a value office. It is not a backup if you have only run one workstream; the interview is a portfolio case and they will smell the gap.

AI PM as a primary is a real option if your last three years included writing specs, sitting in design reviews, and owning a backlog with engineers. It is a weak primary if your proof is "I translated between business and IT." Every delivery manager says that. AI PM hiring managers want a one-pager they can hand a staff engineer: problem, thin slice, eval, risks, open questions. If you cannot imagine writing that tomorrow from the wiki-copilot work, do not pick the seat yet. You can move there after a delivery or FDE year.

Write one sentence why the other three seats are not now. Not forever — now. Example: "Not FDE now: I will not sit in a repo after hours and week 2 confirmed that. Not AI PM now: I have not owned a backlog with engineers; I have owned a RAID. Not Transformation now: I have run one program, not a portfolio of twenty." That sentence is not self-sabotage. It is how you stop applying to posts that will waste a loop. You can revisit in six months with new evidence.

Tell people the pair out loud today. "I am targeting AI Delivery Lead, with Solutions as backup." If that sentence feels too small, you are still addicted to optionality. Optionality is how week 4 becomes a mood board. Hiring managers do not hire mood boards. They hire a person who can say the job they want in the first two lines of a resume and then prove it for forty-five minutes.

Open a post. Ignore the title for sixty seconds. Highlight three buckets: must-have skills you already have from delivery (stakeholder management, RAID, vendors, UAT, steering); must-haves this month gave you (RAG, evals, cost envelopes, HITL, data class); must-haves you still lack (a production repo, a named cloud cert, PyTorch, a published paper). If the post is 80 percent ML research, skip it. If it is 80 percent "drive the backlog with engineering and customers," lean in. You are reading for the work, not for the brand.

A useful scoring pass: five minutes per post, four marks. (1) Building vs delivery vs research — which is the first five bullets? (2) Customer-facing vs internal platform — which room do they describe? (3) Must-haves you can talk about for ten minutes vs must-haves you would bluff. (4) Domain — do you have a right to an opinion (claims, warehouse, banking ops, public sector) or would you be learning the business and the tech at once? If (3) has more than two bluffs, discard. If (4) is a cold domain and (1) is research, discard twice.

Must-haves you still lack are not automatically disqualifying. "Nice to have: Python" on a delivery post is a conversation, not a gate. "Must have: 5 years training production models" is a gate. The skill of reading posts is knowing which line is which. Recruiters copy lists. Hiring managers interview the first bullets. When the first bullets say "partner with customers to scope and land AI use cases, define evals, and run the operating cadence," that is your post even if the requirements blob later says "familiarity with transformers."

Apply to fewer posts than you think. Ten mapped posts today is a lab, not a pipeline. Of those ten, three should be a clear yes for the primary, two a maybe that you would only take as a backup-seat interview, and five a no. If you cannot find five nos, you are not discarding. Discarding is the muscle. Every yes you keep that is actually a research scientist role is an interview you will fail in public and a week of prep you will not get back.

When you cannot search live, invent realistic posts from the Roles page and still do the mapping. Write them as if a real company posted them. "Acme Insurance, AI Delivery Lead, P&C claims. First bullets: stand up an AI factory operating cadence; charter two thin slices; eval plan; vendor POC governance; steer a monthly value committee." Then score yourself. The muscle is the fit logic, not the LinkedIn search syntax. You can search tomorrow. You cannot fake the logic tomorrow if you skip it today.

Keep a running "not for me" list with one line of reason. "Staff ML Engineer, lab X — first bullets are CUDA and training runs." "AI Architect, consultancy Y — first bullets are TOGAF and a multi-year platform, no slice." "Prompt Engineer intern — junior, no delivery surface." This list is how you stop reopening the same posts at 11pm. It is also useful in interviews: "I looked at FDE and at research seats; I am here because your first bullets are the job I have been training for." That sentence is positioning. Most candidates cannot say it because they applied to everything.

Title inflation is real and it is not new. "AI Architect" is sometimes a delivery lead with a nicer badge. "Forward Deployed Engineer" is sometimes solutions with a hoodie. "AI Product Owner" is sometimes a business analyst who sits near a chatbot. "GenAI Lead" can mean anything from a director of a value office to the person who runs the Copilot license spreadsheet. "Prompt Engineer" in 2023 meant a job; in 2026 it often means a task inside another job. Read the first five bullets, not the title.

A noisy title is a delivery problem you already know how to solve. When a program was called "digital transformation" and the work was a vendor cutover, you did not believe the banner. Do the same here. Print the first five bullets. Mark each as Build (write code, train, productionize a model), Deliver (scope, RAID, evals, vendors, change), Sell (discovery, POC, proposal), Product (spec, roadmap, evals as acceptance), or Research (papers, new models, novel training). If four of five are Deliver, it is a delivery seat whatever they called it.

Examples you will see this week. "Staff AI Architect, financial services" whose bullets are: define reference architecture, run design authority, vendor down-select, landing zone, enablement. That is delivery plus solutions, not a compiler job. "FDE, applied AI lab" whose bullets are: embed with customers, stand up RAG, write evals, feed product. That is FDE. "FDE" whose bullets are: run discovery workshops, produce SOWs, deliver executive demos, no mention of a repo. That is solutions. Apply to the work. In the first line of your message, use their title so ATS and the recruiter are not confused, and in the interview speak the work.

Consultancy titles are a special mess. "Managing Consultant, AI" may be a delivery lead. "Senior Consultant, AI" may be a solutions engineer. "Analyst, AI" may be a slide machine. Read staffing: will you be on a client site making a slice, in a factory producing decks, or in a deal room? The same firm will hire all three under similar banners. Ask the recruiter one question before you prep a loop: "What does the first 90 days look like on the account — building, steering, or selling?" If they cannot answer, they are filling a generic req. Treat it as a maybe.

Internal mobility titles are noisy in the other direction. "IT Project Manager — GenAI workstream" may be the best AI Delivery Lead job you will see, with worse pay-band language. "Digital Director" may be Transformation with no AI literacy in the team. Do not skip an internal post because the title is embarrassing. Do skip an external post because the title is glamorous. Your resume headline can say AI Delivery Lead even if the posting says Program Manager, GenAI — as long as the first bullets match and you are not inventing a level.

Write a personal translation table today and keep it next to your applications. When a new title appears, add a row. This is how you stop wasting a Sunday on a "Head of AI Engineering" role that wanted a person who has shipped training pipelines. It is also how you explain a pivot in the screen: "Your posting says AI Architect; the bullets are operating cadence, evals, and vendor POCs. That is the work I do. I use Delivery Lead in my materials so hiring managers in other firms can find me." Clean. No apology.

Allowed: not having trained a model, not being a LeetCode specialist, not having a 10k-star GitHub, not holding a named cloud AI cert, not having a production LLM already in your last company, not having a CS degree. Those gaps will not fail a delivery, solutions, transformation, or most AI PM loops. They will not even fail many FDE loops if you can talk through a thin slice and write a small script. Companies that require all of the above are hiring a different person. Let them.

Fail: cannot talk through a thin slice, no eval story, no opinion on HITL, no cost envelope, cannot tell a specific story from your delivery past, bluff a tool you cannot discuss for ten minutes, cannot say no to a bad use case. This month closed the fail list if you did the labs. If you skipped week 2 and week 3 artifacts, today is the day you notice, not day 29. A loop will not grade your effort. It will grade whether those five fail-items are gone.

The allowed list surprises people who have been on Twitter. They think the gap is Python. The gap is judgment they cannot hear. A hiring manager who has been burned by POCs is not listening for "I fine-tuned Llama." They are listening for "here is the slice, here is how we would know it worked, here is who the human is, here is what it costs at 10×, here is what would make us stop." If you can say that about the wiki copilot and about one real program from your past, you are in the game. If you cannot, a cert will not save you.

There is a third bucket: gaps you will not chase this quarter. Maybe that is a cloud architect cert. Maybe that is becoming fluent in a second language of code. Maybe that is a public GitHub with a polished app. Write them down so they stop leaking into tonight's work. "Will not chase this quarter" is a delivery tool you already use on programs. Use it on yourself. The people who fail this month are not the people with gaps. They are the people who chase eight gaps at once and arrive at day 30 with no pack.

Be precise when you name a remaining gap, because interviews will ask. Bad: "I need more technical depth." Good: "I have not yet run the wiki copilot against a folder of 20 PDFs; that is the next four-week skill if FDE stays primary. I can already design the slice, the eval, and the HITL." The second version is a person you can hire. The first version is a person who will wander. Hiring managers are buying reduced risk. A named gap with a named next step reduces risk. A fog of insecurity increases it.

If a gap is on the fail list and you still have it tonight, do not apply this week. Close it. The wiki design, the charter, the eval table, the cost envelope — those are closable in days, not months, because the academy already asked you to build them. What is not closable in days is a fake production deployment or a fake CS degree. Do not close those by inventing them. Close the fail list with artifacts. Then apply.

Today you map ten posts. You do not apply to ten posts. Mapping is a lab that trains the eye. Applying is a workstream you will install on day 30 with WIP limits. If you apply today, you will apply with positioning v0 and a resume that still says "managed stakeholders and delivered on time." That is how you burn a warm company. Map. Rewrite on day 25. Apply after the pack exists, or apply to one stretch post only if a recruiter is already in your inbox and waiting.

How to pick the ten. Four from your primary title (including noisy variants). Two from the backup. Two that look glamorous but you suspect are research or platform-without-a-slice — you need practice discarding. Two internal or adjacent (same industry, "program manager, data and AI," "engagement manager, digital"). If you cannot search live, write ten realistic posts from the Roles page, including two traps. The output is a table, not a feeling.

For each post write four lines only: company and title as posted; first-bullet digest in your words; fit score 1–5 against the primary; one sentence of reason, including which gap they would probe. Example: "Northwind Bank — AI Delivery Lead. First bullets: operating cadence for 8 use cases, eval standard, vendor POCs, steering. Fit 4. They will probe whether I have killed a use case and whether I can talk cost. Charter covers this; I need a number from a real program in the intro." That is a usable row. "Looks good, will apply" is not a row.

When a post is a 5, steal its language into a keyword list for day 25 — not to stuff, to see which of your real skills they named. When a post is a 1, write the discard reason into the noisy-title table. When a post is a 3, it is usually a backup-seat interview or a title mismatch. Do not treat 3s as 5s because you like the brand. Brand is how people accept loops they will fail. You already know this from programs that were "strategic" and doomed.

Watch for posts that want you to be five seats. "FDE / solutions / PM hybrid, also run the AI office, also code." That is an understaffed startup or a confused enterprise. You may still take the loop if the primary energy matches, but you walk in with a boundary: "I do the delivery and the first slice; I do not also run the platform team." If you cannot imagine saying that, you will be hired into a blender. Delivery managers know blenders. Do not self-select into one as a career move.

End the lab with a shortlist of three companies or post-shapes you would actually want. Write why in one line each, including the room you would be in. "Acme — delivery lead in claims, steering a factory, I have claims-ops proof." "Lab X — FDE, warehouse customers, week 2 energy." "Consultancy Y — solutions, regulated industry, I can qualify." Those three lines become the target list for the job OS on day 30. Without them you will search from scratch every Sunday and call it a strategy.

Positioning v0 from day 1 was allowed to be ugly. Positioning v1 is aimed at one hiring manager for one seat. Structure: one sentence that names the seat, the domain or scale, and the proof. Then ninety words: the messiest program, the decision you owned, the AI-shaped hole you kept hitting, and the muscle this month added (evals, thin slices, HITL, cost — pick two, not five). Stop. If you cannot say it in ninety words, you do not own it yet.

Stealable sentence, delivery primary: "I am a delivery lead who has run multi-vendor operations programs; I am moving into AI Delivery Lead roles where the job is to charter thin slices, put evals on the table, and stop POCs from becoming production by accident." Stealable sentence, FDE primary: "I am a delivery manager who sits in the customer mess and makes a first slice work; I am targeting Forward Deployed Engineer seats where the bar is RAG, evals, and a 30-day implementation walk, not a training pipeline." Stealable sentence, solutions: "I am a delivery lead who already runs discovery and SOWs; I am targeting AI Solutions Engineer seats where the job is to qualify a use case, whiteboard the thin slice, and protect the deal from a promise the product cannot keep."

The ninety words must contain a number from your real past. Headcount, budget, vendors, sites, cycle time, a go-live date, a cut you made. Without a number you sound like every other pivot post. With a number you sound like a person who has run a thing. Then one AI-shaped sentence that is true: "I now design retrieval slices with a golden set, a HITL dial, and a cost envelope at 10× — independent work, not a production claim." That sentence is allowed. "Led enterprise GenAI transformation" is not, unless you did.

Read the paragraph out loud to someone who has seen you deliver, or to a wall. If they cannot tell which seat you want, rewrite. If they hear five tools and no program, rewrite. If they hear apology ("I am just a PM trying to break in"), rewrite. You are not breaking in from nowhere. You have run programs. You are pointing that skill at a new object. The humility that is useful is precision about what you have not shipped. The humility that is useless is erasing the last ten years.

You will paste a version of this into LinkedIn About on day 25 and into the top of emails. That is why it has to survive an honesty pass today. Delete any model name you cannot discuss for ten minutes. Delete "passionate about the future of work." Delete a list of five seats. Keep the program, the number, the seat, and two muscles. Ugly and specific beats polished and generic. You can polish later. You cannot un-smudge a lie in a second loop.

Lock the primary in writing at the top of your notes: "Primary: [seat]. Backup: [seat]. Not now: [the other three], because [one line each]." If you change it after day 25, you redo the resume top third and the 90-second intro. That is allowed, and it costs a day. Changing it every evening because a new post appeared is not allowed. Treat the pick like a scope freeze. You know what happens to programs that refuse to freeze scope. The same thing happens to job searches.