By Janos Barberis
Startup dealflow shows which skills, credentials and entry-level jobs are breaking before official statistics do. Governments that ignore it are flying blind.
Governments track employment statistics to understand how AI is changing work. By the time those numbers move, the change has already happened. Janos Barberis, co-founder and CEO of SuperCharger Ventures, an accelerator for EdTech and Future-of-Work startups, argues there is a faster signal: watch what founders are building. Dealflow shows which skills, credentials and entry-level roles are being disrupted, often years before the shift reaches labour data. Treating it as labour-market intelligence, rather than a niche innovation category, is the cheapest early-warning system a government can get.
Why should governments read startup dealflow as labour-market data?
In 2020, founders used VR to train people for critical jobs during COVID. Training centres could not operate, and VR kept them running. Five years later, it is still scaling. That was the first signal this was not a startup trend and that dealflow was telling something employment statistics were not. The labour market was rerouting itself in real time.
The money confirms it. European funding for learning and work companies more than doubled, from €710 million in 2024 to €1.6 billion in 2025. In the first quarter of 2026, EdTech raised $512 million globally, and more than 70% of that capital was allocated to workforce training, with a focus on employability, certification, onboarding, and upskilling. Even in a cautious market, capital moves to where the skills gap hurts most.
Inside our own pipeline, the shift is starker. Today, 100% of the founders applying to us use AI and large language models as core team members, and many now sell that capability by shipping AI agents to other companies. Three years ago this was not happening at scale. AI went from a feature to a co-worker.
Founders vote with their time before economists vote with their data. When the number of startups attacking a vertical rises quarter on quarter, the pressure behind it rises too. More founders means a bigger, more urgent gap.
Why is the education-to-employment pipeline falling behind?
METR, an independent AI evaluation lab, finds that the length of software and research tasks that frontier AI agents can complete autonomously has roughly doubled every 7 months. A degree takes three or four years. The World Economic Forum expects 39% of core skills to change or become outdated by 2030, which is to say, within a single degree cycle.
The gap is already measurable. In a study across six countries, 53% of employers said they struggle to find graduates with the right AI skills, while 78% of university leaders believed they were meeting employer expectations. Only 14% of graduates rated their own ability to apply AI in a professional workflow as high. Meanwhile, 86% of students already use AI in their studies, and 83% of faculty in DEC’s follow-up survey worry that students cannot critically evaluate what AI produces. Students adopted the tool. Institutions have not converted that usage into workplace capability.
And the pressure lands on juniors first. An analysis of US payroll data found a 16% relative decline in employment among workers aged 22 to 25 in the most AI-exposed occupations. It is early evidence, but it matches what we see in our applications. The routine junior tasks that once served as apprenticeships are exactly what founders are automating.
Which parts of the pipeline break first? Curriculum design and assessment, because they are the easiest targets. Anything that places a person inside a company, such as internships and hiring, is 50/50. Credentials break last, because a credential is mostly a branding exercise, and brands die slowly. China, for example, is not waiting for them to die. Its universities revoked or suspended 12,200 degree programmes between 2021 and 2025, replacing them with AI-aligned majors.
What is the actual human advantage?
Yet, at least now we’re not talking about a full replacement. Around 25% of jobs worldwide are exposed to generative AI, and the most likely outcome is transformation of tasks rather than wholesale replacement. That matches our dealflow as founders today building augmentation, extending the capacity of the existing workforce.
Employers, meanwhile, are misreading the skill. They hire for prompt-writing, a six-month skill that is obsolete before onboarding ends. The scarce skill is judgement, which means knowing which answer to trust and when to override the machine. PwC’s AI Jobs Barometer shows vacancies requiring AI skills grew 69% against 9% for the wider market, with a 62% wage premium, and finds that judgement, leadership and adaptability are now demanded far earlier in careers.
In our portfolio, the most valuable people are not the best prompters. They are the best editors of machine output. They know what is wrong, what is missing, and what to do next. We teach kids to be hard drives, but AI is a perfect hard drive. The human edge is a direction, never retention.
How would an early-warning system actually work?
We can approach it as a radar. While university data tells what was taught and labour statistics show what already broke, founders can tell what breaks next. A government could track three signals every quarter:
| Signal | What to measure quarterly | What it tells you |
| Founder concentration | Change in the number of startups attacking each skills vertical | Where labour-market pressure is building |
| Repeating AI use-cases | Which applications of AI recur across new companies | Which tasks are being redesigned right now |
| Who pays | Whether the learner, the employer or the state buys the product | Where the urgency actually sits |
Overlay the adoption gap between students and institutions, and the dashboard writes itself. None of this requires exotic infrastructure. AI is already used to analyse vacancy data and update qualifications, yet adoption is uneven. 41% of vocational-education policymakers use or know of such tools, against only 21% of education providers. So the main problem is coordination of the data.
The practical agenda for the next months can look like this. Governments should track founders as they do with graduates, so it can be another source of information alongside vacancy data and employer surveys. Then rethink the four-year degree as the primary policy instrument.
Employers should hire for judgement and adaptability. Tools reset every six months, while judgement compounds. AI as infrastructure, so design roles that amplify human bandwidth rather than replace it. Assume every role you post is being redesigned, including the nuone you are hiring for.
Conclusion
Treat EdTech as a niche, and you switch off your own radar. You discover your workforce is obsolete at the same moment your unemployment office does, years too late. The danger is the delay. Nations do not fall behind in AI in one dramatic moment, they fall behind one ignored quarter at a time. By the time the signal reaches the labour statistics, it is an autopsy. And we do not have 20 years to fix this.
