Generative AI Certification Roadmap

A staged path through generative AI credentials — from no-prerequisite fundamentals to LLM engineering and professional GenAI exams — with current 2026 exam facts.

Maya Patel13 May 2026 · 9 min read

Ask AIChatGPT Claude Gemini Copilot Perplexity Grok

Subway-style map of generative AI certification tracks across five providers converging at literacy, builder and production stages

Two years ago, a "generative AI certification roadmap" could not have existed: there was almost nothing proctored to put on it. As of 2026 there is a genuine ladder — AWS launched a professional-level Generative AI Developer certification, NVIDIA and Databricks run associate-level LLM exams, Google Cloud added a business-level Generative AI Leader credential, and Microsoft rebuilt its AI exams around generative and agentic workloads. This article maps that ladder in order: which credential to take at each stage, what each one verifiably costs and covers, and how to sequence them without buying into an exam that is about to change underneath you.

Scope first: this roadmap covers generative-AI-specific credentials — LLMs, prompt engineering, retrieval-augmented generation (RAG), agents, and the platforms that serve them. If you want the broader AI progression including classic ML and data science exams, that lives in our AI certification roadmap; if you want ML-engineering sequencing specifically, see the machine learning certification roadmap. And where this article sequences the path, the ranked comparison of the same credentials lives in Best Generative AI Certifications.

First, know what counts as a certification here

The generative AI training market is flooded with course-completion certificates — Google AI Essentials, DeepLearning.AI specialisations, vendor badges — that are valuable learning but are not proctored certifications. Everything on this roadmap is a supervised, scheduled exam from a recognised provider, because that is what carries weight on a CV. There is currently no dominant standalone "prompt engineering certification" as a proctored exam; prompt engineering instead appears as a domain inside the exams below (it is roughly 30% of Salesforce's Agentforce Specialist exam, for instance, and a named topic in the Databricks Generative AI Engineer Associate guide). If you searched for a prompt engineering cert, the honest answer is: earn one of these instead, and prompt engineering is inside it.

Stage 1 — Generative AI literacy (no prerequisites)

The entry stage proves you understand what foundation models are, what they can and cannot do, and how organisations use them responsibly. None of these exams requires coding.

AWS Certified AI Practitioner (AIF-C01)

The broadest Stage 1 option. As of August 2026: 65 questions, 90 minutes, $100 USD (regional pricing varies), passing score 700 on AWS's 100–1,000 scale, no prerequisites, valid three years. It covers AI/ML fundamentals plus a substantial generative AI and responsible-AI component on AWS services, and it introduced AWS's newer question formats — ordering, matching and case studies — so expect more than plain multiple choice. A useful structural detail: passing the higher Machine Learning Engineer – Associate later auto-recertifies it.

Google Cloud Generative AI Leader

Google's first non-technical generative AI certification, and the purest "GenAI literacy" exam on the list: 50–60 multiple-choice questions, 90 minutes, $99 USD plus tax, proctored online or onsite, valid three years, no prerequisites, aimed at any job role. If your work is strategy, product or management rather than building, this can be your whole roadmap; deeper business-track options are compared in Best AI Certifications for Business Professionals.

Microsoft Azure AI Fundamentals (exam AI-901)

Microsoft's fundamentals certification survived the company's 2026 exam overhaul, but its exam changed: AI-900 was replaced by AI-901, updated for Microsoft Foundry and generative AI, with the certification name unchanged. Fundamentals certifications never expire, passing is 700/1,000, and pricing follows Microsoft's fundamentals tier (commonly cited at about $99 US; it varies by country — confirm on Microsoft Learn). Microsoft does not publish question counts or, for AI-901, an official duration, so distrust any source quoting exact numbers.

Stage 1 verdict: take one, not all three. Choose by the cloud your employer uses; default to AIF-C01 if unsure, because AWS's ladder above it is currently the most complete.

Stage 2 — Hands-on generative AI engineering (associate level)

Stage 2 exams expect you to have built something: a RAG pipeline, an LLM-backed application, prompt workflows against real APIs. None enforces prerequisites, but all assume roughly six months to a year of practice.

NVIDIA-Certified Associate: Generative AI LLMs (NCA-GENL)

The most focused LLM certification available: 50–60 multiple-choice questions in one hour, $125 USD, remote-proctored online (despite persistent blog claims that NVIDIA exams are unproctored — they are not), valid two years with a retake to renew. NVIDIA suggests only a basic grounding in generative AI and LLMs, making this the gentlest Stage 2 entry point. Its sibling exams — NCA Multimodal Generative AI (NCA-GENM) and NCA AI Infrastructure and Operations (NCA-AIIO) — let you extend sideways into multimodal models and GPU infrastructure within the same programme.

Databricks Certified Generative AI Engineer Associate

The strongest platform-anchored option for people shipping LLM applications on a lakehouse stack: 45 scored multiple-choice questions, 90 minutes, $200 USD per attempt, valid two years, offered in English, Japanese, Brazilian Portuguese and Korean. The exam guide (updated March 2026) covers prompt engineering, RAG, vector search, model serving and governance with Mosaic AI. Databricks does not publish a passing score, recommends six-plus months of hands-on generative AI experience, and enforces a 14-day wait between attempts at full price — so arrive prepared. Databricks Learning Festival events, held quarterly, have offered 50% certification discounts.

Microsoft's new AI engineering track (exam AI-103)

Microsoft retired the Azure AI Engineer Associate certification (exam AI-102) on 30 June 2026 and replaced it with AI-103, earning the new Azure AI Apps and Agents Developer Associate credential — focused on generative AI, multimodal models, agentic workflows and Microsoft Foundry. Because the exam is new, its price, length and question count were not officially published at the time of writing; treat it as the Azure-track Stage 2 choice and take the details from Microsoft Learn when you book. Do not spend money on AI-102 study materials — that exam can no longer be taken.

The bridge option: AWS Certified Machine Learning Engineer – Associate

Not purely generative, but increasingly so: AWS announced that the MLA-C02 version (beta registration opening 1 September 2026, replacing MLA-C01, whose last English sitting is 28 September 2026) adds generative AI, agentic AI and foundation-model workloads. At $150 USD, 65 questions and 130 minutes for MLA-C01, it suits engineers who want ML operations depth alongside GenAI. If you are timing a purchase around the version change, the safest instruction is to check AWS's certification page for the current code before booking.

Stage 2 verdict: pick the exam matching your stack — NVIDIA for model-layer depth, Databricks for lakehouse LLM applications, AI-103 for Azure shops, MLA for AWS engineers. One well-chosen Stage 2 credential beats two overlapping ones.

Stage 3 — Production generative AI (professional level)

AWS Certified Generative AI Developer – Professional (AIP-C01)

The first professional-tier certification dedicated to generative AI. It came out of beta in March 2026, validates integrating foundation models into production applications — RAG architectures, vector databases, Amazon Bedrock including AgentCore — and carries no formal prerequisites. AWS's standard professional pricing puts it at $300 USD with three-year validity; secondary sources report 65 scored plus 10 unscored questions and a 750 passing score, but those figures were not confirmed on an official AWS exam page at the time of writing, so verify them in the official exam guide. AWS also launched an "Agentic AI Demonstrated" microcredential alongside it — a lighter add-on, not a substitute.

Stage 3 alternatives barely exist yet; that is the honest state of the market. Specialists in governance and assurance can look at ISACA's Advanced in AI Audit (AAIA) or Advanced in AI Security Management (AAISM), but both demand prior credentials (CISA or an equivalent audit designation for AAIA; CISM or CISSP for AAISM) and belong to audit and security careers rather than a general GenAI build path.

The roadmap at a glance

Stage	Credential	Cost (USD)	Length	Validity
1 — Literacy	AWS AI Practitioner (AIF-C01)	$100	65 Q / 90 min	3 years
1 — Literacy	Google Generative AI Leader	$99 + tax	50–60 Q / 90 min	3 years
1 — Literacy	Azure AI Fundamentals (AI-901)	~$99, varies	not published	no expiry
2 — Builder	NVIDIA NCA-GENL	$125	50–60 Q / 60 min	2 years
2 — Builder	Databricks GenAI Engineer Associate	$200	45 Q / 90 min	2 years
2 — Builder	Microsoft AI-103 (Azure AI Apps and Agents)	not yet published	not yet published	1 year, free renewal expected per Microsoft's associate model
3 — Production	AWS GenAI Developer – Professional (AIP-C01)	$300	reported 65+10 Q (unconfirmed)	3 years

All prices vary by country and tax; confirm on each provider's page before booking.

A worked example: platform engineer to GenAI engineer in three moves

A concrete sequencing scenario. A backend developer at an AWS-based fintech, comfortable with Python but new to LLMs, could run this path over roughly a year:

AIF-C01 first. Cheap, fast, and it forces a structured tour of foundation models, responsible AI and the Bedrock ecosystem before any building begins.
Databricks GenAI Engineer Associate or MLA next, depending on whether her team serves models through a lakehouse or through SageMaker — the certification should chase the production stack, not the other way round.
AIP-C01 once she has shipped. Professional exams reward scar tissue; attempting AIP-C01 before running RAG in production wastes a $300 fee.

Notice what the sequence does at each step: every exam is booked after the corresponding work has started, so the credential certifies something true. That principle generalises to any provider mix.

Mistakes that burn money on this path

Studying for retired or replaced exams. The last 18 months retired AWS ML Specialty (final sitting 31 March 2026), Microsoft AI-102 and DP-100, and replaced AI-900 with AI-901. Any GenAI study plan copied from a 2024 blog post is partly obsolete. Check the provider page, then check it again the week you book.
Collecting Stage 1 certificates. Three literacy badges signal indecision, not depth. One, then build.
Treating course certificates as certifications. A Coursera specialisation teaches; it does not certify in the proctored sense. Recruiters increasingly know the difference.
Ignoring renewal economics. NVIDIA and Databricks credentials last two years and renew only by retake at full or exam price; AWS gives three years and a 50% discount voucher to active cert holders; Microsoft associate certs renew annually but free via an online assessment. Over a five-year career window those models differ by hundreds of dollars.
Sitting the exam cold. Every exam here is scenario-heavy. Work through sample questions per domain, log which domains you miss, and only book once a full timed simulation comes back comfortably above your target. ExamPractice's free sample questions cover exams including the Databricks certification track, and subscribers get timed practice-test simulation for full rehearsals — use the results to target weak domains, not to memorise answers.

Frequently asked questions

Is there a certification just for prompt engineering?

No major provider offers a standalone proctored prompt engineering certification as of August 2026. Prompt engineering is examined inside broader credentials — the Databricks GenAI Engineer Associate guide names it as a topic, and it makes up roughly 30% of Salesforce's Certified Agentforce Specialist exam ($200 since January 2026; the earlier free promotion has ended).

Can I skip Stage 1 if I already build with LLMs?

Yes — no Stage 2 exam listed here has formal prerequisites. Skipping is sensible for working engineers; the fundamentals tier exists for career-changers and for teams that want a shared vocabulary.

How long does the whole roadmap take?

There is no honest universal figure, and providers publish none. The gating factor is hands-on experience: Databricks suggests six-plus months of GenAI practice before its associate exam, and AWS recommends a year or more of applied work before professional level. Plan in quarters, not weekends.

Do generative AI certifications go out of date quickly?

The credentials renew on normal cycles (one to three years), but exam content is being revised unusually fast — Databricks refreshed its GenAI guide in March 2026, AWS is moving MLA to C02 in September 2026, and Google notes ongoing updates. Always study from the current official exam guide, not last year's course.

Choosing your next station on the map

Start from where your work already is. If you are pre-hands-on, book a Stage 1 exam on your employer's cloud this quarter. If you are already shipping LLM features, go straight to the Stage 2 exam that matches your serving stack. And if you have production RAG or agent systems behind you, AIP-C01 is currently the only professional-grade way to certify it — a genuine first-mover credential while the rest of the industry's Stage 3 is still under construction.

Exam facts in this guide were checked against official certification-provider pages on 30 August 2026. Fees, exam codes and policies change — confirm on the provider’s own site before you book.