Certificates Don't Prove Skills: What to Show Instead
NYC banned classroom AI and MIT stopped trusting assignments. Three reports in one week now have the sector hunting the same thing: proof a skill is real.
A course certificate used to answer one question: were you there. In early September, the institutions that issue credentials stopped being sure they can answer even that. On September 1, 2026, New York City, the largest school district in the United States, banned generative AI for students through eighth grade for one year. Six days earlier, findings from an MIT committee, reported on August 26, said that AI can now complete almost all written assignments for its undergraduates.
Neither institution responded with a better certificate. One removed the tool it could not verify. The other stopped trusting the final artifact and started grading the process instead. Both moves land on the same conclusion, and it reaches your resume: proof of skill can no longer ride on proof of attendance.
By early October the rest of the sector had reached the same page, in a single week of reports.
What a certificate ever provedPermalink to “What a certificate ever proved”
A certificate of completion says you enrolled and you finished. It was never a skills test. Its quiet assumption was an honor system: the submitted work was yours. In one analysis of 221 massive open online courses, the median completion rate was 12.6%, so the certificate did prove something rare about persistence. What it never measured was what you can do, and the platforms that sell access have not added a check for that. The career-readiness tools now bundled into subscription learning platforms did not add one either: the ones we checked had no person behind the verdict and no result an employer can open.
Generative AI broke the honor system, not the certificate. When a model can write the essay, pass the quiz, and submit the project, the receipt stops meaning much. That is the shift both institutions reacted to within one week.
New York City chooses the banPermalink to “New York City chooses the ban”
The New York Times reported on September 1, 2026 that New York City schools will bar generative AI for students through eighth grade for the 2026-27 school year, alongside new screen-time limits. The Associated Press, CNN, and the city government confirmed the policy the next day.
A district that runs the largest public school system in the country looked at homework it could no longer trust and chose removal over verification. That is a defensible response for a school, because proctoring every assignment does not scale. It also sets a pattern worth noticing if you learn as an adult. When an institution cannot check presence, its first move is constraint, and a credential issued under constraint says less about the person holding it, not more.
MIT grades the process insteadPermalink to “MIT grades the process instead”
The university answer arrived six days earlier. An MIT committee, in findings reported on August 26, 2026, concluded that AI can complete almost all written assignments for its undergraduates. The committee's recommendations: regular deadlines throughout a project and feedback at each stage, rather than one verdict at the end. It pointedly does not advise relying on AI detectors.
Read that as a hiring document, because it is one. MIT is saying the finished artifact no longer proves authorship, so the evidence has to move into the process: staged work and a person watching it happen. An employer reading your application draws the same conclusion a professor now draws. A polished final piece could be anyone's. A documented, reviewed trail is yours.
The sector ran the same auditPermalink to “The sector ran the same audit”
EdReports, a nonprofit that reviews instructional materials, reported on September 29 that of ten curriculum and edtech providers it studied, only one had third-party evidence that its AI features improve learning outcomes. "I knew that there wasn't going to be much evidence, but I was surprised at how little there was," said Courtney Allison, the group's chief academic officer.
The next day brought usage data. Inside Higher Ed reported on the first analysis of which external tools faculty plug into Canvas, the learning management system that runs much of higher education. The most popular AI tool did not make the top 100. The AI layer is thinner than its marketing.
EDUCAUSE's annual Top 10 priorities list, reported October 2, then put AI in first place for the first time in the list's history, displacing cybersecurity. The top item is "determining where AI adds real value"; the vote drew 882 respondents, the largest pool on record. Number two is future-proofing students for a volatile world. "This doesn't signal consensus as much as it signals urgency," said Mark McCormack of EDUCAUSE.
Take the three together. Vendors ship AI features without evidence they work. In the systems people use every day, the AI layer barely registers. The sector's buyers now rank finding real value as their first priority. So does the employer reading your resume. A badge, like a certificate, is a record of purchase; what every buyer wants is a result they can open.
What proves a skill nowPermalink to “What proves a skill now”
Put the common signals side by side and the hierarchy is blunt:
| Signal | What it attests | Who checks it | Weight with employers |
|---|---|---|---|
| Certificate of completion | You enrolled and finished | The platform that sold the course | Low |
| AI-detector pass | Software guesses a human wrote it | Software | Low; MIT's committee does not advise relying on it |
| AI feature badge on a course | A feature was shipped | No one outside the vendor; 9 of 10 in the EdReports study | Low |
| Artifact built to a spec | You can produce stated work | Anyone who opens it | High |
| Work reviewed stage by stage | The work is yours and improving | A named reviewer | High |
| Public portfolio an employer can open | All of the above, on demand | The employer | Highest |
Three requirements survive that table:
- Artifacts made under constraints. A project with a spec and a deadline, the way real work arrives. Open-ended tinkering proves interest; a fixed brief proves delivery. Our list of portfolio projects that get career changers hired is built on that principle.
- A visible process over time. Staged deadlines with something submitted at each one. This is the MIT recommendation applied to yourself, and it is the only defense that travels: the trail exists whether or not anyone trusts your final file.
- A human verdict. Feedback from someone who can do the work, attached to each stage rather than to the ending. An automated score checks a file; a person checks understanding.
What this looks like on a career switchPermalink to “What this looks like on a career switch”
Start from what you already have. A skills audit maps your current experience against a target role and shows which proofs are missing; the skills audit guide walks through it in five steps, and a free online audit does the inventory in minutes.
If AI tools sit inside your target role, our AI skills assessment narrows the audit to the tasks you can already delegate.
Then build the missing proofs the way MIT grades: one week at a time, each week ending in submitted work. Some certificates still carry weight, mostly proctored or industry-issued exams, and our certificates vs portfolio breakdown covers which ones are worth the money. For everything else, the unit of proof is a finished piece a reviewer has seen.
What proof costsPermalink to “What proof costs”
The money for this pipeline already exists. Workers who pay for their own training spend an average of $788 a year (Careerminds, via HR Executive, August 25, 2026), and our analysis of that week's upskilling data found that time, not price, is the binding constraint. Access subscriptions quietly monetize the gap: with median course completion at 12.6%, a monthly plan gets paid whether anything finishes, as our Coursera vs Udemy comparison works out in dollars.
Traecta is built for the other side of that trade: a weekly plan matched to your target role, every module ending in work you submit, and the work reviewed by a person whose feedback arrives before the next deadline. Finished pieces collect in a portfolio an employer can open without asking you for anything. The paid step follows the rule the sector just rediscovered: a fixed price for a measurable result. A live review of your roadmap and your finished work, thirty minutes with a person, costs $25. The surrounding market sells access instead: content libraries have run 40 to 72 percent off for months, by our weekly price tracking, and none of those prices buys a human looking at your work. Start with the free skills audit; the review only matters once there is work to review.
What to take from this monthPermalink to “What to take from this month”
- A certificate of completion proves attendance, and attendance is now the contested part. NYC banned generative AI through eighth grade for one year on September 1, 2026, and MIT committee findings, reported on August 26, said AI can complete almost all written undergraduate assignments.
- Institutions that cannot verify presence respond with constraints or process checks, not stronger certificates. MIT's fix, regular deadlines plus feedback at each stage and no reliance on AI detectors, is a template you can apply to your own learning this week. A month of sector data agreed: 9 of 10 vendors studied had no third-party evidence for their AI features, the most popular AI tool missed the top 100 of Canvas integrations, and EDUCAUSE's first priority is now finding where AI adds real value.
- The proofs that survive are artifacts built to a spec, a visible process, and a human verdict. Collect all three where an employer can open them.
If your next six months of learning ended today, what could a hiring manager open? Run the free skills audit, see which proofs are missing, and put the equivalent of that $788 a year into the parts that produce evidence.
SourcesPermalink to “Sources”
- The New York Times, September 1, 2026: New York City bans generative AI in schools
- AP, CNN, and nyc.gov confirmations of the NYC policy, September 2, 2026
- Higher Ed Dive, August 26, 2026: AI can complete almost all written assignments for MIT undergrads
- EdSurge, September 29, 2026: EdReports finds only one of 10 EdTech vendors with third-party evidence for its AI features
- Inside Higher Ed, September 30, 2026: first Canvas LMS integration analysis, most popular AI tool outside the top 100
- TechTimes, October 2, 2026: AI tops the EDUCAUSE Top 10 IT Issues list for the first time
- HR Executive, August 25, 2026: workers who self-fund spend $788 a year on upskilling
- Jordan (2015), IRRODL: analysis of 221 MOOCs, median completion rate 12.6%, as cited in our Coursera vs Udemy comparison
