Anyone can use AI, now prove you can be trusted with it

It's a supervised examination with ID and biometric check-in, two 90-minute blocks, in a live sandboxed codebase, scored collaboratively by AI examiners and human experts, with the final call on every result belonging to the human.

There's no question bank to leak, because every sitting generates fresh. The one thing you can't do here is pass on recognition alone.

A hands-on AI certification for software developers. You direct an AI agent through a live codebase, and every prompt you write is part of your answer.

It measures the judgment you wrap around AI, what you send it in to do, what you accept, and what you push back on, in whatever stack you already work in. It isn't proof you can use the tools, it's proof you can direct them.

Nobody has built a certification for that judgment before, which is why this one exists.

No. Competence is the gate, not tenure. The exam proves judgment at whatever level you're exercising it, and the people on our waitlist span the full range of experience almost evenly.

Yours. The exam runs on the stack you already work in, with the AI tooling you already use, Claude Code, Cursor, Copilot, Codex, whatever is open on a normal Tuesday.

Prove judgment in someone else's stack and you're being tested on translation, which is not the thing being certified.

A vendor certification certifies proficiency with that vendor's product, which makes it an adoption funnel with an exam attached. A lab that sells the model can't credibly certify your judgment over that same model's output.

This exam is vendor-neutral by construction, because you choose between pinned frontier models from different vendors inside the sitting itself.

The same logic holds against anything that scores how often you use AI rather than testing what you do with it. Frequency isn't judgment, and using it isn't directing it.

The issuer is the recognition.

• 10,000+ developers and 680+ companies already certify with certificates.dev
• Our exams hold 4.6 on Trustpilot from named, verifiable reviewers, across 100+ reviews
• The Vue track is officially endorsed by Vuejs.org, with Evan You signing every certificate himself
• Every track was built and reviewed by named experts, Google Developer Experts, Microsoft MVPs, and open-source creators

The AI certification adds a score report an employer can actually read, plain trust statements with the evidence linked underneath.

The exam is framework and tool agnostic by construction. If a competent developer from a different stack could reason their way to the right answer, the item is fair.

So what gets scored is your judgment directing AI, not your fluency with whatever interface happens to be popular this month.

It's the division of labor on your certificate, said plainly.

Three AI examiners read every answer independently, from different model families, at a depth no human panel could afford. A board of working developers holds the final signature, and a human expert reviews every fail before it's issued.

The machines bring the scale, the people put their names on the result, and neither alone would be worth printing on paper.

Certificates.dev. The training behind it comes from BitterBrains, which has trained 2,000,000+ developers.

The examination board lead for this certification is Alex Garrett-Smith, Technical Education Lead at BitterBrains. He started recording programming screencasts in 2009 and grew Codecourse to 300,000 subscribers before it was acquired in 2025. He leads the exam's design and facilitates the live cohort.

Alex Kyriakidis, Founder and CEO of Vue School, sits on every certification examination board.

Other named contributors across the tracks include Evan You (creator of Vue.js), Sébastien Chopin (creator of Nuxt), Mostafa Said, Daniel Kelly, Alain Chautard, Aurora Scharff, Steve McDougall and Matteo Collina.

You won't be answering questions. You work inside a real codebase, with history, tests, and a little debt, through a real agentic harness, file edits, a terminal, a plan your assistant can execute.

The people around you are AI agents playing their parts. Your assistant is the AI you direct. The stakeholder wrote the ticket, and the ticket is vague, because real tickets are. The examiner is silent, reading everything you did and how you defended it.

Every prompt you write is part of your answer. The transcript is the answer sheet.

The full walkthrough, ticket to defense, is in the exam section above.

Two supervised blocks of about 90 minutes each, so around three hours in total. Same day by default, or split across a week if that works better for you.

The exam opens on 28 September, and sittings are bookable from then.

You're required to. Directing it well is the entire thing being examined.

A machine you can work on, a webcam, and a stable connection. The sitting is supervised, so tab-switching, developer tools, and a second monitor are all detected and flagged.

Check-in is ID plus a biometric liveness check, then random liveness re-checks through the sitting, so the person who started is the person still in the seat at the end. Keystroke patterns run as a continuous signal in the background.

Every flag routes to a human, and nothing auto-fails you. Full detail on what's collected and how long it's kept lives in the exam handbook, linked here once it's published.

If a technical problem interrupts you, you get a free retake for that session. You don't spend your included retake on our infrastructure failing.

Yes. Appeals get impartial human re-scoring on a published timeline.

Because the transcript makes re-grading a read rather than a re-sit, this is something we can actually offer rather than just promise.

Not a number. A six-domain profile with receipts, every scored claim linked to the exact transcript moment it came from, including where you caught the planted issue and where you missed one. A calibration index shows whether your confidence matched your accuracy.

There's also a hiring-manager view of the same scores, written as plain trust statements like "can be trusted to review agent-written changes before they reach production", built to be read in thirty seconds by someone who wasn't in the room.

A fail gets the same document, which makes it a retake study plan rather than a rejection letter.

Community access, the playbook "Prove Your AI Skills", and your onboarding roadmap land the night you join, not the night the platform opens. The prep guide follows August 24, the platform and your training dashboard September 7, and your cohort starts September 14.

Your trial exam, a full supervised run of the real environment, lands in the run-up to the sitting that counts.

The exam opens a few days before your cohort's final sessions, and sittings are bookable same-day or split across a week, so there's no penalty for waiting until you've finished.

Three weeks live with Alex Garrett-Smith, Technical Education Lead at BitterBrains and Examination Board Lead for this certification.

Six live sessions, two a week, from 14 September. About four hours live per week, plus roughly an hour a day between sessions. Pods of four. Every session recorded and yours to keep.

About 80% of it is hands-on, drill decks, sandbox labs, real codebases, trying the moves yourself under feedback until the judgment is actually yours.

The session-by-session schedule board is in the cohort section above.

Two graded practice sittings. Mock 1 lands in week two and is deliberately harder than the real exam. Mock 2 in week three is a parallel form of it, scored against your day-one baseline.

We grade both of them ourselves. Other programs leave the capstone self-graded, which means you find out how you're doing from the person with the most reason to be optimistic. Here somebody else grades you, and both sittings are debriefed, so you walk into the exam knowing where you stand.

Every session is recorded and yours to keep. A rough time zone costs you the live room, not the material.

The self-paced side doesn't depend on your calendar at all. The playbook and community land the night you join, the training platform follows on 7 September, and you work through it at your own pace.

The only September-bound piece is Cohort 1 itself. If it doesn't fit, you can take a seat in a later cohort, live, when one does.

A self-guided platform that opens on 7 September, a week before the cohort begins. It holds every format the exam itself uses, at your own pace.

• Judgment decks, MCQs and real-world case exercises that grow your personal calibration curve with every rep
• Stakeholder-interrogation sims, a vague ticket across the table from you until interrogating it is a habit
• Transcript-grading drills, reading an agent's work the way the exam's silent examiner does
• Planted-trap review labs, diffs seeded with real defects, dialed from obvious to nearly invisible

You solve every case yourself first, then see exactly how we'd have solved it.

Your training dashboard opens with the platform and starts keeping score, your calibration curve, mock results, trial exam result, and your certificate once you've earned it, all in one place.

A guide released on 24 August that maps the six domains the exam grades. Inside:

• Worked exam-style items for each of the six domains
• The scoring mechanics written out in full, call semantics and the C1 to C3 confidence matrix
• A trap-taxonomy field guide naming the failure patterns AI reliably ships, so you recognize them on sight
• A calibration primer, teaching you to size your confidence against real stakes before you declare a level that gets scored
• A study plan carrying you from the day you open it to the morning you sit

No. Your trial exam is a full supervised run of the real environment before the sitting that counts, same two-block structure, same live codebase, no certificate attached, just the format. The first time you see it isn't the day it counts.

Being first is the part no later buyer can ever have. Your name goes permanently on the Cohort 1 roster, your cohort dates are the earliest there will ever be, and you certify before the general public can even start.

Certificates carry a Cohort 1 marking and a serial number assigned in order.

Yes. Everyone will eventually be able to get certified.

Later cohorts cost more, start later, and train with the same live system, just without the seat at the very front. What won't come back is this price, these terms, and being first in the room.

It's the second of the two purchase options. Everything in the Cohort 1 seat, plus a full year of access to every other certification we run.

That's every track up to Mid level for 12 months, across Vue.js, Nuxt, React, Angular, JavaScript and Laravel, each with its own exam and self-paced training bundle. On top of that you get one Senior certification in the track of your choice, a retake on every exam, and TypeScript the day it launches.

The 12 months start from your purchase date, and every one of those certifications is already live, so there's nothing to wait for. Email team@certificates.dev with the framework you want and we'll set you up the same day.

Nobody redeems all of them, and nobody needs to. Two certifications at today's listed prices already cost more than the upgrade.

Yes. We can upgrade your account, and you'd cover the difference between what you already paid and the All Framework Access price. The rest of the certifications open up right away.

Email team@certificates.dev from the address you purchased with, and we'll confirm the amount and make the change.

It runs from the moment you buy until 30 days past your cohort's start date, not 30 days from purchase. That's deliberate. You'll have sat through real live sessions before the window even begins to close, so you're deciding on the thing itself, after you've seen it.

Sitting an exam ends it. That's the rule we've always run, you can't take the exam and then ask for the money back, and on All Framework Access it applies to any exam in the year, not only the AI one.

To request a refund, email team@certificates.dev. A refund happens from a single message.

One retake is included with every purchase, and the pass assurance makes it automatic. Sit the exam, miss, and your retake plus a seat in the next cohort's prep are yours, no case to make.

Failing doesn't trigger the money-back guarantee, it triggers this instead, and it holds whether or not your refund window is still open. A miss is covered either way.

The score report you get is the study plan for the second attempt. The exam is designed to be passed by capable engineers who prepare.

No. Nothing shows on your profile, and there's no page anyone can look up to see that you sat the exam and didn't pass. The attempt is kept private so the bar can stay high.

You get the difference back, no asking required. If we ever offer this training and certification bundle for less, the price honor covers you automatically.

Even if you're a senior developer, ten years in and already using AI every day, you'll leave measurably more skillful or your money back. Earlier in your career, the gap we close is bigger, not smaller, and the guarantee is the same.

Measurably isn't a figure of speech. Day one is a measured baseline and you sit an exit measurement on the same instrument, so the delta is something you can see.

Yes. Buying now locks the best price this certification will ever have, and the self-paced side doesn't depend on your calendar. The playbook and community land the night you join, the training platform follows on 7 September, and you work through it at your own pace.

The only September-bound piece is Cohort 1 itself. If it doesn't fit, you take a seat in a later cohort, live, when one does.

The dates are committed and shown again at checkout, and the refund runs to 30 days past your cohort's start, which means you hold a full-refund option well after you've seen the real thing. We priced the promise so that we only win if we deliver.

Credit card, debit card, Apple Pay, Google Pay, and PayPal, through Chargebee's secure checkout.

For teams paying by bank transfer, email team@certificates.dev.

Yes. A US sticker price isn't the same real cost everywhere.

If you're browsing from one of about 134 eligible countries, you'll see a regional pricing checkbox above the pricing cards, ticked by default, and it applies to both tiers. Unticking it reverts to the global rate for comparison.

The discount is tied to attending your cohort and sitting the exam from that same eligible country, not just buying from it.

Yes, 40% off on student verification. Email team@certificates.dev with either a copy of your student ID or a message from your student email address, and we'll apply the discount.

Yes. Add your VAT number, EIN, or company details to your invoice after purchase, from your account settings.

3 years from your purchase date, the same as our other certifications. There's no rush, you can sit whenever you're ready inside that window.

English. That's the only language it runs in today.

Yes. Seats can be added directly on either pricing card and the price adjusts as your team grows, with volume rungs from 5 seats. All Framework Access is available per seat too.

For a larger rollout, procurement requirements, or anything a toggle can't handle, email team@certificates.dev and we'll work it out with you directly.

We already wrote the note to your manager, linked right under the pricing cards. Send it as is or edit it first. It makes the case with the exam's trust statements, "can be trusted to review agent-written changes before they reach production", which is a different conversation than "I've been using AI a lot lately".

If budget approval won't clear before the window closes, email us from your company address and we'll hold your price for 14 days while it runs.