# Hiring AI engineers: work ownership review kit

This is a role-design worksheet, not a candidate assessment. It contains no profiles, employee ratings, scoring, hiring prediction or software integration. Use it without a Metix AI account. All application/platform team examples are invented. The separate public-role read-through paraphrases one employer listing observed November 5, 2026; it is not that employer's hiring rubric or an endorsement.

## Start from work, then choose the role

1. Put your actual AI product or production problem above the blank map in a separate note. Decide whether you need an ongoing employee, a bounded project or an existing team dependency. These files do not recommend a staffing model or estimate demand.
2. For each workstream, describe the work to deliver. Name its current owner; leave unknown when you cannot verify one. Describe what the new hire would own, contribute to, or leave outside the role. One hire does not automatically replace every team in the model lifecycle.
3. Confirm each dependency with the technical owner. For example, a feature engineer may depend on a platform team for inference capacity. An unknown platform owner is an unanswered design question, not proof you need an infrastructure expert as well.
4. Write evidence that would let an experienced engineer discuss relevant past work. These are employer preparation notes, not interview questions or mandatory tests. Use your existing interview-scorecard process once responsibilities are agreed.
5. Review the map with the engineering lead. Transfer the agreed boundaries into your existing hiring brief. Record changes and unresolved questions in the log. The CSV does not create a job, search profiles, evaluate people or send outreach.

## Compare the invented examples

The application example concerns one document-assistant feature. It puts product integration, task evaluation and application triage in the proposed hire's scope. The platform team still owns shared serving infrastructure. Base-model training and novel research are outside this invented project.

The platform example concerns operating shared inference infrastructure. It puts serving reliability and platform triage in the proposed hire's scope; product teams retain their own application integration and task-quality criteria. External workshops are not automatically part of the role. Neither example is a universal AI job taxonomy.

Compare the model_serving and production_incidents rows first. If those ownership decisions change, the evidence you need changes too. A long list containing every AI framework would miss that distinction. Do not copy these ownership allocations into an unrelated role.

## Read a real title cautiously

public-role-readthrough.csv uses OpenAI's public Applied AI Engineer, API Platform, Digital Natives listing. The listing places applied work in a customer-facing technical context. Not stated means the reviewed posting does not establish that responsibility or credential; it does not mean the employer rejects it or that the capability is unnecessary. Check the live page before relying on a location or qualification. The original listing may change or close.

No complete job posting, compensation figures, personal data or third-party prose is included. We paraphrased only the points needed for this limited source-reading demonstration. Do not use this employer's seniority, location or practices as a general benchmark.

## File and review checks

Every map has eight named workstreams and seven columns. The public-role table has six rows with explicit source sections and a separate unknown state. Open CSVs in a spreadsheet or plain-text editor; retain the headers. Verify each dependency with its owner and preserve a versioned copy. A blank field stays unknown; it is never automatically treated as an exclusion or a new must-have.

This kit resolves work scope. It is separate from general intake forms, candidate shortlists, search-string generators and interview scorecards. Those tools can take over once the employer has agreed what the role actually needs to deliver.