| Path | Core work | Closest BLS occupation and May 2025 median pay | Portfolio project |
|---|---|---|---|
| AI evaluation | Test sets, rubrics, grading, inter-rater agreement | Data scientists, $120,230 | A reproducible evaluation with an agreement check |
| AI workflow design | Task mapping, guardrails, measuring time saved | Operations research analysts, $88,940 | One reviewed workflow with before-and-after timing |
The two paths
AI evaluation
AI evaluation answers whether a model, prompt or product does a job well enough, with evidence another person can check. The work covers representative and adversarial test sets, rubrics written before results, a choice of code-based, model-based or human grading, and agreement statistics such as Cohen's kappa. The AI evaluation learning path walks through each step with free resources, including the open-source Inspect framework developed by the UK AI Security Institute and Meridian Labs.
- Best for: people with domain expertise, careful writing and basic statistics.
- Main limitation: there is no official occupation with this name, so titles vary widely.
- Pay context (checked 16 October 2026): the BLS data scientists page lists a $120,230 median annual wage for May 2025 and 35% projected growth from 2025 to 2035.
AI workflow design
AI workflow design turns "use AI for this" into a documented process with inputs, permissions, review points and recovery steps. The work covers mapping tasks by volume and error cost, choosing the simplest design that works, specifying guardrails and human handoffs, and measuring time saved including review. The AI workflow design learning path lists free guides from Anthropic, OpenAI, Google and NIST.
- Best for: people from operations, project management or analyst roles who know how work actually gets done.
- Main limitation: claimed time savings are easy to inflate, so credibility depends on honest measurement.
- Pay context (checked 16 October 2026): the BLS operations research analysts page lists a $88,940 median annual wage for May 2025 and 12% projected growth from 2025 to 2035.
The US Bureau of Labor Statistics lists a $120,230 median wage for data scientists and $88,940 for operations research analysts in May 2025.
Why these two skills
Use of generative AI is now common at work, which makes evaluation and process design more valuable than tool tricks. St. Louis Fed researchers reported in November 2025 that 37.4% of US adults aged 18 to 64 used generative AI for work by August 2025, and estimated time savings equal to 1.6% of all work hours. Organizations that want more than that average need people who can test whether a tool is good enough and redesign the work around it safely.
Both paths also connect to risk management. The NIST AI Risk Management Framework, released on 26 January 2023 for voluntary use, covers trustworthiness across the design, development, use and evaluation of AI systems. Knowing its vocabulary helps when your work meets legal, security or compliance teams.
A portfolio that shows one honest, reproducible result beats a list of tools you have tried.
How to use these paths
- Pick one path based on the work you enjoy, not the title you want.
- Choose a small, real task from your current job, a volunteer role or a public dataset.
- Follow the path's steps and keep notes as you go: versions, dates, decisions.
- Publish the write-up, including failures and limits.
- Ask someone else to rerun it. If they cannot, fix the documentation.
Keep the first project small enough to finish. A focused evaluation of one task with 50 to 100 examples, or one workflow with 30 timed cases, gives another person something they can rerun, and that rerun is what makes the result credible. Record review and correction time alongside any time saved, because a figure that leaves them out overstates the gain.
For background on reading published scores critically, see our guide to reading AI benchmarks.
Looking for paid work now
These guides build skills. To find open roles, internships and paid AI training work, start at our AI jobs desk, which lists current AI internships, AI training jobs with platform pay rates and AI engineer salary ranges by role, each checked at the employer's or platform's own page.
All six career paths
- Machine learning engineer career path
- Prompt engineering career path
- AI product manager career path
- AI ethics and governance career path
- AI evaluation career path
- AI automation and workflow design career path
How we checked this
We read the BLS pages for data scientists and operations research analysts, the NIST AI RMF page, the St. Louis Fed's November 2025 post and the Inspect documentation on 16 October 2026. Pay figures are national medians for related occupations and do not describe any specific AI job title.
Frequently asked questions
Are these paths job listings? No. They are editorial learning guides that end in a portfolio project.
Which path is easier to start without coding? Workflow design, if you already know a business process well. Evaluation also welcomes domain experts, but basic Python helps you run evaluations at scale.
Do I need a degree? Not for learning the skills. BLS lists a bachelor's degree as typical entry-level education for both related occupations, and some employers prefer more.
How long does a portfolio project take? It depends on scope. A focused project on one task with 50 to 100 examples, or one workflow with 30 timed cases, is a realistic first goal.
Can I do both paths? Yes. Evaluation skills make workflow designs more credible, because you can prove the AI step is good enough before you rely on it.
Are the resources really free? The official guides and documentation linked in each path are free to read. Some vendor courses and certificates may charge, so check before you enroll.