I work for equity, not fees. That means I don’t charge by the hour or take a retainer; I take a stake in the companies I advise, so my success depends on theirs. On LinkedIn, the list of startups and companies I’ve worked with has passed 100. The practical consequence is that I don’t see AI through conference keynotes. I see it in Monday morning conversations: who we hire, what we automate, where we got burned.
In April 2023 I wrote here that GPT-4 was not the end of jobs. Two years have passed. I won’t name companies or share internal numbers, because they aren’t mine to share. But I can tell you about the patterns that keep repeating, almost regardless of industry.
What happened in public over the last few months
A few moments moved the debate from theory to practice. In April 2025, Shopify’s CEO told employees in an internal memo that was later made public that teams must show why AI can’t do a job before asking for more headcount. Also in April, Duolingo’s CEO announced an “AI-first” approach, including gradually stopping using contractors for work AI can handle, and faced a fairly harsh public backlash.
The most instructive case, to me, is Klarna. In early 2024 the company said its AI assistant was doing the work of about 700 customer-service agents. In May 2025 its CEO said Klarna would invest in human customer service again, so customers can always talk to a person. That isn’t a failure of the technology. It’s a course correction, and it’s exactly the kind of lesson I see on a smaller scale.
On the big-picture numbers, the World Economic Forum’s Future of Jobs Report 2025, published in January, estimated around 170 million jobs created and 92 million displaced globally by 2030, a net gain of roughly 78 million. These are estimates, not prophecies. But their direction matches what I see: more churn than disappearance.
What I see in the companies I work with
- Junior hiring for repetitive roles has slowed. First drafts, basic support, simple data work: exactly the tasks that used to be your way into a profession are now done acceptably by a tool. Companies aren’t necessarily firing people, but they’re not opening those positions. For young people, this is the most worrying signal in the whole picture.
- Small teams do more. A startup that a few years ago needed ten people to reach a working product can often manage with fewer today. Not because people work longer hours, but because a lot of the filler work has gone.
- Roles are being redefined, not wiped out en masse. The marketer writes less and edits more. The analyst spends less time cleaning data and more time interpreting it. The job title stays; what’s inside it changes.
- Companies want people who combine domain expertise with AI. The hardest person to find isn’t the “AI expert”. It’s the accountant, lawyer or salesperson who knows their craft deeply and also knows how to put AI to work in it.
Where I see things going wrong
The most common failure pattern is the manager who announces “from tomorrow we all use AI” with no training, no process and no clarity on who checks the output. The team gets a subscription and a vague expectation. The result: people pretending to use the tool, people using it badly, and a boss convinced that “AI doesn’t work”. Last autumn I wrote about what AI taught me about productivity: the gain comes from changing how you work, not from installing an app.
The second pattern is in customer support. Automation works well for questions that keep repeating. It works much better when a customer can reach a human quickly if the problem is complicated or they’re upset. Companies that shut the door on people paid for it in trust. Klarna showed this at scale; I’ve seen it at small scale.
Third: quality control and taking responsibility are becoming core skills, not details. When a model produces in five minutes what a person used to produce in a day, the value shifts to whoever can say “this is wrong” and sign off on the result. I wrote recently that accountability stays with people, and that holds inside a company too, not just between citizens and the state.
What I got right and wrong in 2023
Rereading the April 2023 piece, I think the core idea held up: tasks get automated, not whole professions, and “AI won’t take your job, a colleague who knows how to use AI will” is truer every month. The list of exposed tasks was pretty good too: repetitive writing, first drafts, basic support, simple code.
What I didn’t see clearly enough was who would pay first. I was thinking of 50-year-olds whose jobs disappear. The more immediate problem turned out to be at the other end: young people who can no longer find the first rung of the ladder. And I underestimated how quickly companies would build AI into their hiring policy, not just their workflow.
Advice for employees
- Be the person who brings AI into the team, not the person it happens to. Propose what gets automated yourself.
- Go deeper into your field. AI without domain knowledge produces text that is polished and wrong.
- Learn to verify: sources, numbers, logic. Everyone will pay for that skill.
- If you’re just starting out, look for roles that give you access to experienced people, not just tasks.
Advice for founders
- Don’t mandate “use AI” without training, time to learn and clear rules on data and review.
- Measure outcomes, not the number of prompts. If a process hasn’t improved, you haven’t automated anything.
- Automate support with a fast path to a human. Customer trust costs more than a salary.
- When you put AI agents that take actions into real processes, decide in advance who answers for their mistakes.
Juniors and the social contract
I want to be blunt here. If every company cuts junior roles because a tool does the work acceptably, in five years we won’t have seniors. Seniors don’t fall from the sky; they’re former juniors who made mistakes on small stakes, under the eye of someone more experienced. A company that drops juniors saves money today and pays dearly tomorrow.
I think we need a new social contract, and not only at government level. Companies gaining productivity from AI should reinvest part of it in training young people, even if the first year isn’t “profitable”. Schools and universities should teach AI as a professional tool, not treat it only as a temptation to cheat. And governments should back real internships and apprenticeships. For years I’ve told young people, including in my advice for young entrepreneurs, to learn fast and find mentors. Now it’s our turn, those of us with experience, to make room for them.
What are you seeing where you work? Is your company still hiring juniors, and if so, what does their first month look like now?