what companies hire ai engineers? A buyer's glossary

How to use this glossary for nearshore hiring in Latin America

Start with the work you need someone to finish. A company adding an AI search feature needs a different interview from a company training a model. Both may advertise for an AI engineer.

Write the expected result before choosing the title. For example: "Build a search feature that returns answers from documents the user is allowed to read." That gives you something concrete to discuss with candidates. It also exposes questions about access controls and testing before interviews begin.

Use the terms below to check job descriptions and hiring proposals. If two people use a term differently, replace it with a written responsibility. "AI training included" should lead to a description of the course and assessment. "Remote" should lead to an agreed work location and schedule.

A technical manager should own the assessment of engineering work. The hiring buyer should check the service terms and cost. Give both people the same role description so a commercial promise does not become an assumed technical qualification.

Keep a short decision sheet beside each proposal. Record what the hire will own, who will manage them, and which party employs them. Add the working hours and costs you have agreed. Leave room for questions that must be settled before an offer.

When speaking with Opus, bring that role description. This site is an Opus-owned reference; recommendations are editorial.

Talk to Opus about a role.

Which terms should you check before choosing a hire?

Separate training people from training models

"AI training companies" can describe businesses teaching employees to use AI tools. It can also describe work involved in training AI models. Decide which service you need before comparing proposals.

For employee training, ask what the person must do after the course. An engineer might need to review AI-generated code and explain why a suggested change is unsafe. Request the scored work behind a course completion badge.

For model training, define the data and intended behavior. Ask candidates how they would test the result against the current approach. Give them room to explain why training a model may be unnecessary for the problem.

Fine-tuning means further training an existing model. Inference means using a trained model to produce an output. Put the right term in the job description so candidates know whether they will change model behavior or build software around it.

For example, a support tool might use an existing model without further training. Assess the candidate's ability to connect it to your application and handle failed requests. If the role includes training, ask how they choose examples and keep test data separate.

Translate job titles into ownership

A role description should tell the candidate what they will own and who makes final technical decisions. "Senior AI engineer" leaves both questions open.

For an application engineer, describe the feature and its users. For a machine learning engineer, state which model work belongs to the role. For an MLOps role, describe the release process and support duties. Make combined responsibilities visible before asking for a price.

An AI-fluent software engineer may use AI tools to write or review code. Assess their ability to check that code, explain it, and maintain it. Avoid treating tool familiarity as proof that someone can train a model or design your whole AI system.

Set the boundary between the new hire and your existing team. Who approves architecture changes? Who can release software? Who answers a customer when an AI feature gives a wrong answer? Put these decisions in the role scope before interviews begin.

If you are replacing a departing engineer, review the work they actually owned. Their title may conceal support duties or knowledge that your new hire will need.

Turn evaluation into a hiring task

Evaluation means checking a system against defined expectations. In a hiring process, ask candidates to show both the test and the reasoning behind it.

Use the same task when comparing candidates for the same role. Decide in advance whether AI tools are allowed. If they are, assess how the candidate checks the output. Ask them to explain a bad suggestion as well as a working solution.

Keep the exercise small enough to review carefully. Use sample data and a clear scoring guide. Avoid giving candidates customer records or asking them to complete unpaid production work.

For an AI search task, include a question the documents can answer and one they cannot. Add a document the test user should not access. Ask the candidate to explain what the system should return in each case.

Score the work against the duties you plan to assign. A polished demo should not cancel out a failure to enforce access rules. Record the reason for each score so the hiring manager can compare evidence after the interview.

Check what technical vetting and certification cover

Technical vetting is the set of checks used to assess a candidate. Certification is a credential tied to a program. Ask for the details behind each label.

Request the assessment criteria and the candidate's results. Check whether the task involved the language or systems your team uses. Ask who reviewed the work and whether you can discuss it with the candidate.

For a training program, request the course scope and passing requirements. Find out whether the person completed an exercise or only attended lessons. Then choose a follow-up question that connects the training to your role.

For example, ask the candidate to review a generated code change that includes a subtle permissions error. Have them explain the defect and propose a check that would catch it. That gives your technical lead something specific to assess.

Keep training plans separate from hiring requirements. If you expect to teach a tool after the person joins, say so. If they must already know it, test that skill before the offer.

Check who owns the work and who employs the person

A recruiting service, an employment service, and a managed project address different buyer needs. Read the proposed responsibilities before comparing fees.

Direct placement describes a recruiter finding someone the client hires. Staff augmentation describes adding people to an existing team. An employer of record is the legal employer. A managed service assigns the provider a defined service or deliverable.

These labels still need a written scope. Ask whether the agreement includes candidate search and technical screening. Identify who signs with the worker and who handles payroll. Confirm who answers employment questions during the engagement.

If you want someone on your engineering team, identify their manager. State who reviews code and decides what ships. If you want a provider to deliver a project, define what you will accept and who fixes rejected work.

Ask about the end of the engagement before it begins. Assign ownership of account removal, equipment return, and the transfer of unfinished work. Review these duties with the people who will carry them out.

Turn time zone overlap into calendar hours

Nearshore location helps you discuss schedules, but a country name does not tell you when someone will work. Record agreed hours in the hire's local time and your team's time.

Mexico City stays at UTC-6 without seasonal clock changes. Tijuana follows US daylight saving time. Bogotá stays at UTC-5, while Buenos Aires and São Paulo stay at UTC-3. These city differences affect the schedule you put in a job description. IANA North America data, IANA South America data.

Check the schedule when your US team changes its clocks. Distinguish meeting overlap from support coverage. A shared afternoon for code review does not establish evening incident response.

Ask the candidate to confirm which recurring meetings they can attend. Put on-call duties in the role description before the offer. State how the team handles a handoff when colleagues finish at different times.

Remote work also needs an agreed location policy. Ask whether the person may work while traveling and who approves a location change. Have your security and employment reviewers check the proposed arrangement before promising flexibility.

Compare salary, monthly rate, and total cost

A salary and a provider's monthly invoice measure different things. Ask each provider to identify included costs, possible extra charges, and the terms for a rate change. Compare proposals for the same role and working arrangement.

Use the same currency and billing period on your comparison sheet. Keep recurring charges separate from setup fees. Ask whether equipment and software seats sit inside the quote or remain your responsibility.

A low headline amount is difficult to budget around if the scope is unclear. Request a written explanation of what happens during leave and when the engagement ends. Ask which charges continue while a replacement search is underway.

Avoid comparing a junior application role with a senior role that owns model development and production support. Match the responsibilities before deciding which quote costs less.

Once the scope is settled, have the person who approves the budget review it. They should see the same duties and schedule that the candidate sees.

Keep shortlist, hire, and start dates separate

A shortlist date tells you when candidates reach your team. An accepted offer tells you someone has agreed to join. The start date tells your manager when to prepare onboarding.

Ask each provider what starts and stops its hiring clock. Check whether interview scheduling time falls inside the estimate. Agree who gives feedback and how quickly they can review candidates.

Build your internal schedule around decisions you control. Reserve interview time before candidates arrive. Choose who can approve the offer and who steps in if that person is away.

Before the start date, assign an owner for account access and the first task. Choose work that lets the hire learn the codebase without carrying an urgent release alone. Give the manager time to review the result and answer questions.

For an AI role, list approved tools and data access rules during onboarding. Tell the hire where to ask before sending company information to a new service.

Read the replacement guarantee before relying on it

A replacement guarantee describes a contractual remedy when a placement needs replacing. Ask what triggers it and what the provider must do next.

Check whether the remedy is another search, a fee credit, or something else. Read exclusions and notice requirements. Ask who pays ongoing costs while the search runs.

Plan for continuity separately. Name the person who will maintain the code and answer urgent questions during a gap. Keep runbooks and access records current while the original hire is still working.

Before approving a proposal, bring the definitions together in a written scope. It should name the work, the manager, the employing party, the agreed hours, and the cost. Ask for unresolved terms to be settled before you choose a finalist.

Talk to Opus about a role.

<!-- Not saved to disk because this session has read-only filesystem access. Rendered-page checks were not run. -->