Certifications
What's real, and what's a badge
The AI certification gold rush is here, and it looks exactly like the Salesforce one did. Same LinkedIn banners, same badge stacks, same quiet suspicion that most of it means nothing. Some of it does mean something. Here is how to tell the difference in about thirty seconds.
I spent years inside Salesforce certification culture, including a fair stretch being part of the problem: talking about certs, counting certs, treating the badge row as a CV. It took an embarrassingly long time to notice that the people I most wanted on a project were rarely the people with the most badges, and that the correlation sometimes ran the other way.
That same wave is now breaking over AI, faster and with less quality control. I have done three of the OpenAI programmes myself. So this page is not sniping from outside, it is a note from someone who bought in.
What actually exists right now
Live from the vendors who publish a machine-readable catalogue, refreshed daily. This is the honest half: full data where a vendor publishes it, and an availability check where they do not. We do not scrape credential lists out of marketing pages, because that breaks quietly and being wrong would be worse than showing less.
Two absences worth noticing. The tools consultants actually spend their days in mostly do not certify anyone. There is no Cursor certification, no xAI or Grok credential, no formal programme for most of the agentic workbench. Meanwhile Anthropic's four exams are properly proctored through Pearson VUE, which puts them above every completion badge here, and all four require a company email tied to a Claude Partner Network organisation. Both things are true at once: technically real, and a channel credential by design. That is the fifth question in the framework doing its job.
The one question that settles most of it
Before frameworks and tier lists, there is a single filter that sorts perhaps eighty percent of credentials correctly.
Can it fail you? If completing the material guarantees the credential, it is a receipt for attendance, not an assessment. That is not automatically worthless, but it is not evidence of competence and should never be presented as though it were. A test you cannot fail measures one thing: that you turned up.
The five questions
When the first one is ambiguous, these settle it. They work for any vendor, any subject, and they will still work when the current crop of credentials has been replaced.
Is there a pass mark, a proctor, a time limit? Unlimited retakes with the same question bank is a formality, not a filter.
Does it ask you to build, debug or decide, or to recall which menu holds which setting? Recall expires the moment the product ships a new release.
A vendor certifying its own product has an interest in you passing, and a bigger one in you buying. That does not make it fake, but read the incentive.
Is there a registry a hiring manager can check without asking you for a screenshot? Verifiable beats impressive.
Some programmes exist to make you capable. Others exist to make a partner channel sell more. Both are legitimate. They are not the same thing and should not share a word.
The tiers, graded honestly
Categories rather than a leaderboard, because individual exams change faster than any page can track. Find the shape your credential fits and grade accordingly.
Proctored vendor exams
Real signalInvigilated, timed, with a genuine failure rate and a verification portal an employer can check independently. These are the ones that get a CV past a filter, and the market prices them accordingly.
Shape: the AWS, Microsoft Azure and Google Cloud AI exams. Proctored, vendor-verifiable, renewable.
Worth doing if the platform is genuinely in your stack. Worthless as a substitute for having built something.
Practical or portfolio-assessed credentials
Real signalRarer, harder, and the most honest form of the thing: you are assessed on work you produced, by someone qualified to judge it. Expensive to run, which is exactly why so few exist.
If you find one in your field, it beats every badge on this page.
Foundational vendor certificates
Useful, modestEntry-level exams that establish vocabulary. Genuinely useful when you are new, or when a client wants to see that the person in the room knows what a context window is. The mistake is presenting one as expertise.
Fine as a first step. A problem when it is the only step, repeated across six vendors.
Completion badges from video courses
Near-zero signalWatch the modules, receive the badge. Nobody fails. A credential anyone can generate in an afternoon is indistinguishable from one earned with real effort, which means it signals nothing to the person reading your profile, and they know it.
Learn from the course if it is good. Just do not wear it as a qualification.
Partner and channel enablement
Not what the name suggestsProgrammes that train partners to position, qualify and sell a vendor's product. This is a completely legitimate thing to build, and the good ones teach real delivery discipline along the way. The problem is purely nominal: several are branded with words like "practitioner", which in every other technical field means someone who can do the work.
You can usually spot one from the first module title. If it opens on market positioning and opportunity routing rather than on the technology, you are being enabled to sell, not certified to practise.
Do them if you are a partner. Do not put them on a CV as evidence of technical capability, and be wary of anyone who does.
What Salesforce already taught us
The AI ecosystem is about to relearn this the expensive way, so it is worth stating plainly.
Salesforce built the most comprehensive certification programme in enterprise software, and for a while a cert genuinely was a differentiator. Then the incentives did what incentives do. Certification became a partner-tier requirement, so partners pushed staff through them in volume. Badge counts became a LinkedIn genre. Braindumps became an industry. And the signal decayed, not because the exams got easier, but because the correlation between holding the badge and being good at the job got weaker every year.
The lesson is not that certifications are worthless. It is that a credential's value is inversely proportional to how badly the ecosystem wants you to have one. When a vendor needs 300,000 certified people by December, the thing they are optimising is not rigour.
The part where I argue against myself
Because the cynical take has a real cost, and pretending otherwise would be dishonest.
Sneering at certifications is a comfortable position for someone who already has a network, a track record and people who vouch for them. If you have none of those, a certification may be the only lever available. It is cheap relative to a degree, it is available from anywhere, and it does not require knowing the right people. Plenty of good practitioners started with a badge that got them a first interview they would never otherwise have had.
So the honest position is narrower than "certs are rubbish". It is this: a certification is a door, not a destination. Use one to get read. Use the work to get hired. The cringe was never in holding certifications, it was in mistaking the collection for the competence, and in an industry that encouraged exactly that because it sold seats.
What to do instead, or alongside. One proctored exam in the platform you actually use, then stop. Put the remaining time into something a stranger can inspect: a repository, a write-up of a migration you survived, a small tool that solves a real annoyance. Our own wreck-watch exists because a problem was irritating enough to fix, and it has opened more conversations than any badge ever did. See the Kit Bag for what practitioners actually run, and the careers guide for the longer route.
Grade your own
Run the five questions over whatever is currently in your LinkedIn banner. If a credential cannot fail you, was issued by someone who profits from you holding it, cannot be verified by a stranger, and exists to help a channel sell, it is a badge. Keep it if you like it. Just do not let it do work it cannot do.
Disagree? Say so.
This is one practitioner's opinion, formed in a different ecosystem and carried into this one. If you have hired for AI roles and found the opposite, that is genuinely more interesting than agreement. Tell us and we will publish the correction.