Certifications Worth Having

The idea in one minute#

A certification proves that you studied a vendor’s platform and passed a test about it. It does not prove you can run a discovery workshop, defend a cost model or say no to an executive. Employers know this. They use certifications for two things: as a filter when screening applications, and — at vendors and partners — because a firm needs a certain number of certified staff to keep its partner status.

So treat certifications as a door-opener and a study plan, not as the destination. The sensible set for an AI solution architect is small: one cloud architecture certification on the platform you will work on, one AI certification on the same platform, and optionally one from a model provider. Three well-chosen credentials and a portfolio beat nine badges.

The certification landscape was rebuilt during 2025 and 2026 around generative AI and agents. Details below were checked on 13 October 2026 against the providers’ own pages; exams, prices and names change, so confirm before you book.

A picture#

flowchart LR
  F[":i-book-open: <b>Foundational</b><br/><small>vocabulary; optional if you have experience</small>"] --> A[":i-cloud: <b>Cloud architecture</b><br/><small>one platform, associate or professional</small>"]
  A --> G[":i-brain: <b>AI on that platform</b><br/><small>generative AI and agents</small>"]
  G --> X[":i-layers: <b>Architect-level AI</b><br/><small>agentic architecture, expert tier</small>"]
  G -.-> M[":anthropic: <b>Model provider</b><br/><small>optional; shows depth</small>"]
  P[":i-file-text: <b>Portfolio</b><br/><small>design doc, cost model, eval plan</small>"] -.->|"what actually gets the job"| X
  class F neutral
  class A queue
  class G compute
  class X io
  class M memory
  class P warn

How it really works#

Amazon Web Services#

The Generative AI Developer – Professional exam is 75 questions in 180 minutes, costs $300 and needs a scaled score of 750 out of 1,000. AWS weights its five domains as: foundation model integration, data management and compliance 31%; implementation and integration 26%; AI safety, security and governance 20%; operational efficiency and optimisation 12%; testing, validation and troubleshooting 11%. The intended candidate has two or more years building production applications and about a year of hands-on generative AI work.

Microsoft#

AI-103 is a 120-minute exam. The AB-100 description is worth reading even if you never sit it: its list of competencies — design agentic-first solutions, orchestrate multiple agents, secure models and data flows, monitor agents, “conduct a return-on-investment analysis of an AI-powered solution” — is a compact job description for an enterprise AI architect.

Microsoft role-based certifications expire yearly and are renewed free with an online assessment.

Google Cloud#

Google notes that its exams are being updated for the product changes announced at Cloud Next 2026, including the move from Vertex AI to the Gemini Enterprise Agent Platform. Use current study material.

Model providers and others#

Choosing your three#

Pick by where you will work, not by which badge is most fashionable.

One cloud, properly, is the rule. The three platforms share the same concepts, and an interviewer would rather meet someone who knows one deeply than someone who has memorised three sets of service names.

How to study so that it is not wasted#

• Build while you study. For every exam domain, build the smallest thing that exercises it in a free-tier account. The exam then tests memories of things you did.
• Use the official exam guide as the syllabus. Every domain and its weight is published.
• Use the provider’s free training first — AWS Skill Builder, Microsoft Learn, Google Skills, Anthropic Academy. Pay for practice questions only at the end.
• Time-box it. Six to eight weeks per exam alongside a job. Longer and you forget the start.
• Write down what surprised you. Those notes are interview material.

What a certification will not do#

Cost and upkeep#

Budget for the whole set, not one exam: fees of roughly $100–$300 each, renewal every one to three years, and thirty to sixty hours of study apiece. Many employers pay; vendors run regular voucher and discount programmes. If you are paying yourself, the associate architecture certificate is the one with the clearest return.

Remember this#

• Certifications open doors and structure study. They do not prove judgement.
• Three is enough: cloud architecture, AI on the same cloud, optionally a model provider.
• One cloud, properly. Choose it by where you will work.
• The 2026 catalogue is new: AWS AIP-C01, Microsoft AI-103 and AB-100, Google’s Agentic Architect, Anthropic’s Claude Certified Architect.
• Build something for every exam domain. Pair every badge with a portfolio artefact.

Try it#

1. Choose your cloud and write down your three certifications in order, with a target month for each.
2. Download the official exam guide for the AI exam on that cloud. Mark each domain as “have done”, “have read” or “new”. The “new” list is your study plan.
3. Read the AB-100 certification description. Which of its listed competencies could you demonstrate today with something you have written or built?

Check yourself#

1. What are the two real uses employers make of certifications?
2. Why should an ML engineer take an architecture exam before an AI exam?
3. Which two domains carry the most weight in the AWS Generative AI Developer – Professional exam, and what does that tell you about where to spend study time?
4. Name three things the job requires that no certification demonstrates.

Sources#

• AWS Certified Generative AI Developer – Professional and its exam guide
• AWS Certification
• Microsoft Certified: Azure AI Apps and Agents Developer Associate
• Microsoft Certified: Agentic AI Business Solutions Architect Expert
• Google Cloud certifications, Generative AI Leader and Professional Agentic Architect
• Claude Certified Architect – Foundations