Some titles are durable role families. Others are specializations, capabilities, or qualifiers. Find the scope that fits your work.
Where you sit on these two axes defines which role fits you.
You work closer to model behavior: training, fine-tuning, evaluation, and safety. The roles vary from research-heavy to production evaluation.
You build systems that ship. Models are a component inside product, platform, security, governance, or customer-deployment work.
This map describes ownership boundaries, not market share. A current vacancy confirms title usage at one employer and date, not the size or durability of the wider market.
Start with durable role families, then inspect narrower specializations and capabilities.
22 profiles mapped
Role families describe a durable ownership boundary supported by current employer postings.
Specializations narrow the work inside a broader engineering or product role.
Capabilities are skills teams need, but they rarely justify a dedicated seat by themselves.
Title qualifiers describe context such as company stage, not a separate discipline.
Design and test instructions for reliable model behavior inside broader product, evaluation, and domain roles.
Design systems that give models the right information, at the right time, in the right format.
Build end-to-end AI applications, from model integration and evaluation to production monitoring.
Turn available models into production behavior for a defined product or workflow.
Specialize in model integration, adaptation, fine-tuning, and evaluation infrastructure.
Design agent systems that plan, use tools, and execute multi-step tasks under explicit controls.
Own the AI core of an early-stage company across architecture, product delivery, and customer feedback.
Own system-level decisions for enterprise AI architecture, security, governance, and integration.
Build the internal platform that makes AI development reliable, secure, observable, and reusable.
Build constraints, feedback loops, and knowledge systems that keep coding agents productive.
Own AI products, including requirements, evaluation criteria, risk tradeoffs, and non-deterministic UX.
Reduce harmful or unintended model and system behavior through safeguards, testing, and risk controls.
Develop, train, deploy, and maintain machine-learning models and their production pipelines.
Operate model delivery, observability, drift detection, and deployment infrastructure.
Connect an AI platform with developers through demos, education, community work, and product feedback.
Connect AI capabilities to enterprise systems, data sources, and operational workflows.
Write testable specifications that agents can use to plan, implement, and validate changes.
Design authentication, delegated authority, permission scoping, and auditability for AI agents.
Build the continuous measurement layer for AI quality, regressions, and production behavior.
Own deployment of an AI system inside a customer environment, from discovery through adoption and handoff.
Protect AI applications, models, data paths, agents, and platforms against misuse and attack.
Turn AI policy, risk, and regulatory requirements into operational tooling and evidence.
A broader title is credible when your portfolio shows the missing ownership, not when the label changes first.
Starting from
Evidence to buildBuild production AI evidence
Starting from
Evidence to buildAdd backend, eval, and AI integration evidence
Starting from
Evidence to buildAdd discovery and rollout ownership
Starting from
Evidence to buildAdd model operations and observability
Starting from
Evidence to buildAccumulate system-level ownership
Starting from
Evidence to buildAdd ML and evaluation foundations
Starting from
Evidence to buildAdd evaluation and AI UX practice
Starting from
Evidence to buildExtend AppSec, cloud, and IAM into AI systems
Starting from
Evidence to buildBuild operational controls and evidence
These artifacts demonstrate production judgment. They do not replace domain, research, security, or regulatory depth where the role requires it.
The evidence shows how titles and skills are used. It does not prove a single AI-driven cause.
Role evolution
The data supports broader labels and ownership, not the disappearance of front-end work.
Stack Overflow's 2025 role dataset reports 26.96% full-stack and 4.25% front-end among 43,560 answers to its role question. In France, the figures are 25.21% and 3.48%.
This is respondent self-identification, not a count of vacancies. It cannot prove that AI caused the difference.
LinkedIn's 2026 US talent report keeps React and JavaScript among leading skills of recent software-engineering hires, while Python, cloud, and AI demand also rises.
The practical positioning is Software Engineer or Product Engineer with front-end depth. Keep Front-end Engineer when browser architecture, accessibility, interaction quality, and performance remain the ownership boundary.
Using an AI coding agent does not by itself justify an AI Engineer title.
Compensation evidence
One posting establishes one employer's range for one role, level, location, and date. It does not establish a global average or a premium caused by the title.
$197K to $278K base
OpenAI, Enterprise, US. Equity separate. Employer-specific customer-facing scope.
$162K to $280K base
OpenAI, San Francisco. Equity separate. One location and a broad level range.
$131.54K to $197K base
Marvell, Santa Clara or Austin. Employer-specific enterprise cyber-engineering role.
Postings observed during the August 31, 2026 review. Verify that each vacancy remains open before using its range.
Continue in the guide
Each role has a complete profile: detailed responsibilities, required skills, entry paths, and where the role is heading. Free, open-source.
I do AI engineering consulting: agent architecture, Claude Code adoption for teams, and getting agentic workflows from demo to production.
