Five seats a delivery manager can actually enter

Who hires this

Consulting firms, enterprise PMO / digital offices, and product companies standing up an AI factory. Closest leap from a classic delivery manager seat.

Why a delivery manager

The job is still delivery: scope, RAID, vendors, steering, change. The delta is that the workstream now includes models, evals, data readiness, and a new class of operational risk. You already run the meeting. This month teaches you what must be on the agenda.

A day in the life

Morning standup across a mixed squad (data scientist, engineer, change lead, business owner). Midday: vendor checkpoint or eval review. Afternoon: steering pack — value, risk, cost per task, go-live criteria. You are the person who translates 'the model is 73% on the golden set' into a release decision.

Gaps this month closes

• You can challenge a technical design without bluffing
• You can write an AI RAID log that is not a copy of a waterfall RAID log
• You can cost a workload and stop a POC that cannot scale

Interview shape

Case + stakeholder. Expect: 'A business unit wants a chatbot by Friday. What do you do in week one?' They are scoring judgment, sequencing, and whether you protect the company from a demo that becomes production by accident.

After 30 days

You can walk into an internal mobility conversation or an external 'AI delivery' posting with a use-case charter, an eval plan, and a vocabulary that survives a hiring manager who has been burned by POCs.

Who hires this

Palantir-style, Databricks, OpenAI, Anthropic, and AI startups that embed people at the customer to make the product actually work.

Why a delivery manager

FDEs are hired for the combination you already have — sit with a messy client, find the real problem, ship something in their environment — plus enough technical depth to 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.

A day in the life

On-site or in the customer tenant. You map a process, pull a slice of data, stand up a thin slice (retrieval + a prompt + a human review queue), measure it, and expand. You are part consultant, part builder, part product manager. The calendar is customer workshops, not internal standups.

Gaps this month closes

• You can design a RAG / agent thin slice and talk through the failure modes
• You can write the first eval set with the customer, not after go-live
• You have a portfolio narrative: problem → slice → measure → expand

Interview shape

Technical screen (Python + LLM API comfort, not LeetCode-heavy at many firms) plus a customer scenario: 'Warehouse ops wants an agent. The data is a mess. Walk the first 30 days.' Some loops include a take-home: a small RAG or a written implementation plan.

After 30 days

You will not be a staff engineer. You will be a delivery leader who can build the first slice and not get lost when the conversation turns to embeddings, tools, or evals — which is exactly the FDE hiring bar at many firms.

Who hires this

Model labs, vector-DB and orchestration vendors, cloud AI practices, and anyone whose product is sold into enterprises that 'need a demo that looks like their world.'

Why a delivery manager

You already discovery-workshop, handle objections, and write statements of work. Solutions is that job with a technical spine: you design the reference architecture on a whiteboard, scope the POC, and protect the deal from a promise the product cannot keep.

A day in the life

Discovery call, then a scoped POC. You pull sample documents, wire retrieval, put a thin UI in front of a champion, and present results with caveats. Between calls you write the technical appendix of the proposal: data, SSO, VPC, evals, cost.

Gaps this month closes

• You can whiteboard RAG vs fine-tune vs agent and pick one with a reason
• You can size tokens and dollars so a POC does not become a surprise invoice
• You can say no to a use case in a way that still advances the relationship

Interview shape

Whiteboard architecture + a mock discovery call. They want to hear you qualify, not pitch. A good answer names data sensitivity, evals, and the human in the loop before it names the model.

After 30 days

You can run a 45-minute discovery and leave with a written use-case score, a proposed thin slice, and a list of blockers — the artifact a solutions interview is secretly testing for.

Who hires this

Product companies adding copilots into existing suites, and internal 'AI product' teams that sit between platform engineering and the business.

Why a delivery manager

You already sequence work, cut scope, and manage stakeholders who want everything. AI PM adds a new product surface: prompts, evals, and model behavior that drifts. Delivery managers who have shipped with engineering, not just reported on it, convert well.

A day in the life

PRDs that include eval metrics, not just user stories. Office hours with design and research on failure cases (the model was confident and wrong). Prioritization against a cost and latency budget. You spend as much time on 'when should the product refuse?' as on 'what should it do?'

Gaps this month closes

• You can specify an AI feature as a contract: inputs, tools, evals, refusal policy
• You understand why 'just add a chatbot' is not a roadmap item
• You can sit in a model-choice conversation and ask about evals, not vibes

Interview shape

Product sense + AI-specific: 'Design an inbox copilot for account managers.' Strong answers start with jobs-to-be-done, then evals, then the model. Weak answers start with GPT-whatever and a prompt.

After 30 days

You can write a one-pager for an AI feature that a staff engineer would respect: problem, thin slice, eval, risks, open questions. That document is your portfolio.

Who hires this

Large enterprises building an 'AI office,' and consultancies staffing transformation programs. Often a director-track seat.

Why a delivery manager

This is portfolio management with a new technology. You already know how programs die: no owner, no data, no change plan, no benefits tracking. AI offices fail the same way, faster, because every VP has a chatbot idea.

A day in the life

Intake of 40 ideas, scoring, a kill list, and a funded set of 5. You run the operating cadence of the AI office: use-case reviews, platform decisions, risk, value realization. You spend political capital saying no.

Gaps this month closes

• A scoring model that is not 'HIPPO wanted it'
• A language for feasibility that includes data, evals, and change — not just 'we have Azure'
• A 90-day value story that finance will not laugh at

Interview shape

Executive case: 'We spent a year and have 12 POCs and no production. Diagnose and reset the operating model.' They want a sequence, a governance design, and the courage to stop work.

After 30 days

You can put a one-page AI operating model on the table: intake, score, thin-slice, eval, scale or kill — and talk to it for 30 minutes. That is the interview, and it is also the job.