AI Certification Roadmaps by Role (2026)
By Kyle Tice · Updated October 19, 2026 · Facts checked against official exam pages
The right AI certification path depends on the job you want, not on which exam is most popular. Pick your role below. Each roadmap starts with free learning, adds one inexpensive exam to prove the basics, then moves to the one or two certifications employers in that role actually ask for. Every path also includes a portfolio step, because for most technical roles the project matters more than the certificate.
All prices are exam fees in USD as of October 2026, taken from our certification database. Timelines assume 5–10 hours of study a week.
How to read these roadmaps
- Steps are in order. Skip a step if you can already pass its exam comfortably, but don’t skip the portfolio step.
- Pick one cloud. Where a step lists AWS, Microsoft and Google options, choose the one your target employers use. Doing all three wastes money.
- Stop when you’re hired. Later steps are for promotion and specialization.
Not sure which role fits? Try the which-cert tool.
1. Developer or AI engineer
AI engineer is #1 on LinkedIn’s 2026 Jobs on the Rise list, and AI engineering is the top skill on its 2026 Skills on the Rise list. The work is building LLM apps: RAG, agents, tool use over MCP, evaluation and guardrails.
Free starting point: Anthropic Academy (Building with the Claude API, Intro to MCP, Claude Code in Action), the Hugging Face Agents Course and the Hugging Face MCP Course. DeepLearning.AI short courses fill gaps quickly.
| Step | What | Cert id | Cost | Time |
|---|---|---|---|---|
| 1 | Free courses above | anthropic-academy, hf-agents-course |
$0 | 3–4 weeks |
| 2 | Cloud associate developer exam (pick one) | ms-ai-103 or databricks-genai-engineer |
$165–$200 | 8–12 weeks |
| 3 | Portfolio: RAG app plus an agent with MCP tools and an eval harness | — | $0–$50 in API costs | 4–6 weeks |
| 4 | Vendor-neutral protocol cert | lf-mcp-associate |
$250 | 4–6 weeks |
| 5 | Professional exam (pick one) | aws-genai-developer-pro or gcp-agentic-architect |
$200–$300 | 10–14 weeks |
Total: about $615–$750 and 7–10 months.
Notes:
- AWS has no associate-level GenAI developer exam, so AWS developers can go straight to AIP-C01 once they have hands-on Bedrock experience. Consider AIF-C01 ($100) first only if you are new to AWS.
- If you work at a Claude Partner Network firm, the Claude Certified Architect – Foundations ($125) fits at step 4. It isn’t open to individuals outside partner organizations. See the Claude certification guide.
- Google’s Agentic Architect is valid for only 12 months, so time it to a job search.
Skip certs entirely if you already ship LLM features at work. A public repo and a write-up of your eval results will do more than any exam.
2. Data scientist or ML engineer
This path is about training, deploying and operating models. GenAI content now appears in every ML exam, but the core is still data, training pipelines and MLOps.
Free starting point: Kaggle Learn, CS50’s Introduction to AI with Python, Andrej Karpathy’s Neural Networks: Zero to Hero, and the official PyTorch tutorials.
| Step | What | Cert id | Cost | Time |
|---|---|---|---|---|
| 1 | Free foundations | kaggle-learn, harvard-cs50-ai |
$0 | 6–10 weeks |
| 2 | Framework cert | lf-pytorch-associate |
$250 (one free retake) | 6–8 weeks |
| 3 | Portfolio: fine-tune and evaluate a small model; build an MLOps pipeline | — | $0–$100 GPU time | 6–8 weeks |
| 4 | Cloud ML engineer exam (pick one) | aws-ml-engineer-associate, gcp-ml-engineer or ms-ai-300 |
$150–$200 | 10–14 weeks |
| 5 | Specialize (optional) | databricks-ml-professional or nvidia-nca-genl |
$125–$200 | 6–10 weeks |
Total: about $400–$650 and 8–12 months.
Notes:
- The AWS ML Specialty retired on March 31, 2026, and MLA-C02 is its successor. MLA-C02 is in beta until it goes live, and AWS allows only one attempt during a beta.
- DP-100 retired on June 1, 2026. AI-300 replaces it. Microsoft’s role-based certs need a free online renewal every 12 months.
- Google’s PMLE is one of the most respected cloud ML exams, but recommends 3+ years of industry experience. Don’t make it your first exam.
See our AWS MLA vs Google PMLE comparison.
3. Product manager or business role
You need enough AI literacy to scope use cases, judge vendors and talk with engineers. You don’t need to code.
Free starting point: AWS Skill Builder’s AI Practitioner prep, Google Cloud Skills Boost’s Generative AI Leader path, and Anthropic Academy’s AI Fluency courses. Oracle’s AI Foundations and Agentic AI Foundations exams are free and proctored, which makes them a zero-cost first credential.
| Step | What | Cert id | Cost | Time |
|---|---|---|---|---|
| 1 | Free courses, optional free Oracle exam | oracle-ai-foundations |
$0 | 2–3 weeks |
| 2 | Cloud AI fundamentals (pick one) | aws-ai-practitioner, gcp-genai-leader or ms-ai-901 |
$99–$100 | 3–6 weeks |
| 3 | Portfolio: a one-page AI use-case brief with ROI, risks and an evaluation plan | — | $0 | 1–2 weeks |
| 4 | Strategy exam | aws-ai-business-strategist |
$100 ($50 beta) | 3–5 weeks |
| 5 | Delivery methodology (optional, for program managers) | pmi-cpmai |
$899 | 4–6 weeks |
Total: about $200 for steps 1–4 and 3–4 months. Add $899 if you take CPMAI.
Notes:
- AI-901 now expects you to read basic Python, unlike the AI-900 it replaced. If you want zero code, pick AIF-C01 or Generative AI Leader.
- The AWS AI Business Strategist exam is in beta. Passing by February 15, 2027 earns an Early Adopter badge.
Skip expensive leadership programs unless your employer pays. For PMs, a strong use-case brief or prototype demo is the better signal. See the AIF-C01 vs Google GenAI Leader comparison.
4. Governance, legal or compliance
Demand here comes from the EU AI Act, ISO/IEC 42001 and customer due diligence. See the full AI governance certifications guide.
Free starting point: the IAPP AIGP Body of Knowledge (free download), the NIST AI Risk Management Framework, and the EU AI Act text. Anthropic Academy’s AI Fluency series covers how LLMs behave in practice.
| Step | What | Cert id | Cost | Time |
|---|---|---|---|---|
| 1 | Free frameworks reading | — | $0 | 3–4 weeks |
| 2 | Technical literacy (optional but useful) | aws-ai-practitioner or gcp-genai-leader |
$99–$100 | 3–5 weeks |
| 3 | Core governance cert | iapp-aigp |
$799 ($649 member) | 6–10 weeks |
| 4 | Portfolio: an AI risk assessment and a model card for a realistic system | — | $0 | 2–3 weeks |
| 5 | Specialize (pick one if relevant) | iso-42001-li or isaca-aaia (requires CISA) |
About $650–$800 | 4–8 weeks |
Total: about $800–$1,700 and 5–8 months.
Notes: AIGP has no prerequisites. AAIA requires an active CISA (or CIA, US CPA or ACCA with an IT audit focus). Only take the ISO 42001 Lead Implementer if your organization is actually pursuing 42001.
5. Security
Security teams are being asked to threat-model LLM apps, red-team agents and write AI security policy. See the AI security certifications guide.
Free starting point: the OWASP Top 10 for LLM Applications, MITRE ATLAS, and the Hugging Face MCP Course (to understand how agent tools are wired before you try to attack them).
| Step | What | Cert id | Cost | Time |
|---|---|---|---|---|
| 1 | Free reading plus security fundamentals (Security+ if you lack it) | — | $0 | 3–6 weeks |
| 2 | Practitioner AI security cert | comptia-secai |
About $298–$359 | 6–8 weeks |
| 3 | Portfolio: prompt-injection test suite and a guarded agent | — | $0–$30 | 3–4 weeks |
| 4a | Management track (requires CISM or CISSP) | isaca-aaism |
$599 ($459 member) | 4–6 weeks |
| 4b | Hands-on track (employer-funded) | giac-gaips or GIAC GAIPT |
$999+ with SANS course extra | 6–10 weeks |
Total: about $900–$1,000 for steps 1–4a and 4–6 months. The GIAC route costs much more because of SANS training.
Notes: ISC2 has announced an AI security certification with a pilot planned for late 2026, but it can’t be booked yet. GIAC GAIPT goes on general sale December 15, 2026.
6. Career switcher
If you are moving into AI from a non-technical field, certifications help you get interviews, but a portfolio gets you hired. Expect this to take a year or more.
Free starting point: CS50’s Introduction to AI with Python (free certificate), Kaggle Learn’s Python and ML courses, the Hugging Face Agents Course, and roadmap.sh’s AI engineer roadmap.
| Step | What | Cert id | Cost | Time |
|---|---|---|---|---|
| 1 | Learn Python and AI basics | harvard-cs50-ai, kaggle-learn |
$0 | 10–14 weeks |
| 2 | First proctored credential (free or cheap) | oracle-ai-foundations (free) or aws-ai-practitioner |
$0–$100 | 3–5 weeks |
| 3 | Structured program (optional) | ibm-ai-engineering |
About $294 (6 months of Coursera) | 4–6 months |
| 4 | Portfolio: two or three projects with public repos and write-ups | — | $0–$100 | 2–3 months |
| 5 | One associate exam that matches your target job | ms-ai-103, databricks-genai-engineer or aws-ml-engineer-associate |
$150–$200 | 8–12 weeks |
Total: about $150–$600 and 9–15 months.
Notes: Lightcast found that 51% of AI-skill job postings are outside IT. Your previous field is an asset. A RAG app over legal, medical or finance documents built by someone who knows the domain stands out more than a generic chatbot. If you are leaning toward governance rather than engineering, switch to roadmap 4 after step 2.
Cost and time summary
As of October 2026:
| Role | Exam fees (typical) | Timeline | Must-have cert |
|---|---|---|---|
| Developer / AI engineer | $615–$750 | 7–10 months | AI-103 or Databricks GenAI, then AIP-C01 |
| Data scientist / ML engineer | $400–$650 | 8–12 months | MLA-C02, PMLE or AI-300 |
| PM / business | About $200 | 3–4 months | AIF-C01 or GenAI Leader |
| Governance / legal | $800–$1,700 | 5–8 months | AIGP |
| Security | $900–$1,000 | 4–6 months | SecAI+ |
| Career switcher | $150–$600 | 9–15 months | One associate exam plus a portfolio |
Cut costs with current vouchers and the free AI certifications list. The cost calculator includes renewal fees, which add up for annual-renewal certs like Microsoft’s role-based exams and Anthropic’s.
FAQ
Should I get a fundamentals cert first?
Only if you are new to AI or new to that cloud. Experienced developers and data scientists can usually go straight to an associate exam. Fundamentals certs carry little weight on an experienced engineer’s resume.
AWS, Azure or Google Cloud?
Check 20 job postings in your target city and count. Pick the one that shows up most. If it’s a tie, pick the one your current employer or your strongest network contacts use. Read our cloud study guides for AWS, Microsoft and Google Cloud.
How many AI certifications do I need?
Usually two: one that proves the basics on your platform and one at associate or professional level. More than three starts to look like collecting. Spend the extra time on projects.
Are certifications worth it compared with a portfolio?
For technical roles, the portfolio usually matters more. Certifications matter most in regulated industries, consulting and partner firms, and for getting past resume filters. See are AI certifications worth it.