I. The confession
For nine years I ran a product consultancy. For nine years I wrote job descriptions asking for unicorns.
Designers who code. PMs who do SQL. Marketers who run paid, write the copy, and design the landing page. Founders who can ship the MVP, file the trademark, and reconcile the books before bed.
I told myself I was hiring rare talent.
I was hiring people to do four jobs because I couldn’t afford four people.
The polite name for it was “T-shaped.” The honest name was “structurally under-resourced and calling it strategy.”
I’m not alone in this. The product community has been calling out this exact pattern for years. It just never made it into the way most of us actually hired.
II. The myth was never about talent
Art Leyzerovich’s widely-shared piece on hiring the Product Management unicorn lays it out clearly: hiring teams pile every conceivable requirement into a single CPO or VP Product role — market strategy, deep technical chops, P&L ownership, organisational design, customer empathy, executive presence — and then act surprised when nobody fits. And the ones who do get hired are gone in 2–3 years, because the role is structurally impossible to do well.
Pendo’s framing is sharper. Hiring managers aren’t really looking for unicorns. They’re looking for “purple squirrels” — the perfect candidate who also works for peanuts. The fix, Pendo argues, was never finding them. It was building teams of T-shaped experts, each with one deep specialty and a general understanding of adjacent disciplines.
Sometimes the unicorn doesn’t exist because you created it. Stack eight years of experience, three certifications, leadership history, and a specific industry background into a single job description, and you’re not describing a hire. You’re describing a myth.
That is RecruitBPM’s point, and it is the bluntest of the three.
And the structural critique — the one nobody quite wanted to say out loud — was that early-stage startups didn’t have the headcount or the budget for separate roles. So they bundled two jobs into one job description, gave it a senior title, and called the requirement “rare talent.”
It wasn’t rare talent. It was rare budget.
The villain of this story isn’t AI. It isn’t hiring managers. It isn’t candidates. It’s the collective lie we all participated in — that breadth, in the absence of resourcing, was somehow excellence.
III. The reveal
Then AI showed up. And here’s where most LinkedIn posts get it wrong.
The dominant narrative is: “AI makes everyone a unicorn now.” That’s the easy take. It’s also wrong.
The sharper read is this: AI didn’t make us all unicorns. It exposed that most of what we called “unicorn skills” was mechanical translation work between disciplines. And translation work is precisely what AI eats for breakfast.
The unicorn UX designer who “also codes” — Claude, Cursor, v0, Lovable will take a spec and produce production-grade frontend. The translation from design intent to working component used to be the rare skill. It is now a tool feature.
The unicorn PM who “also writes SQL” — every modern analytics tool ships natural language query. The translation from “I have a question about user behaviour” to “here is the SQL that answers it” used to be the rare skill. It is now a tool feature.
The unicorn marketer who “also designs” — Midjourney, Canva AI, Figma AI. The translation from copy concept to visual asset used to be the rare skill. It is now a tool feature.
The unicorn founder who “also handles legal and ops and finance” — Harvey for legal review, an army of bookkeeping agents, automated compliance tooling. The translation work was the bottleneck. The translation work is now automated.
And the structural evidence backs this up.
Then the case studies arrived.
And Dario Amodei gave the one-person billion-dollar company 70–80% odds for 2026, at Anthropic’s developer conference in May 2025.
The picture isn’t “AI made everyone a unicorn.” The picture is much stranger. AI revealed that what we called “unicorn breadth” was often work that didn’t need a human at all. And the people we labelled as exceptional were, in many cases, just doing the translation labour the organisation couldn’t afford to specialise out.
IV. The shadow side
This is where most posts about AI productivity stop. With the Medvi number. With the Base44 headline. With the line about taste.
That isn’t the honest version.
The honest version is that the 10x productivity claim is partly a lie. A UC Berkeley Haas study, published in Harvard Business Review in February 2026, followed 200 employees at a tech company over eight months as they adopted AI tools. The headline finding: AI tools didn’t reduce work. They intensified it. Workers took on a broader scope of tasks, worked at a faster pace, and extended work into more hours of the day. Several participants noted they felt more productive but did not feel less busy — and in some cases felt busier than before. Output volume up. Output quality down. Worker satisfaction down. Boundaries between work and home blurred.
The same study found that engineers were now spending time reviewing and correcting “vibe-coded” output from non-engineering colleagues — which the Berkeley researchers described as a downstream cost of the productivity narrative.
The solo unicorn comes with its own hidden costs.
Medvi’s customer service chatbot, in its early deployment, fabricated drug prices that Gallagher then had to personally honour. It hallucinated product lines that didn’t exist. The founder remained the only human backstop for every system failure, at any hour, at any scale. Maor Shlomo, at an EY conference shortly after the Base44 acquisition, was candid:
Yes, absolutely I felt lonely. When the difficult moments come, there’s no one to share the burden with.
The Ardent VC framing of the “10x knowledge worker” sharpens the point: the leverage isn’t in using AI. It’s in orchestrating it. Engineers running five Cursor agents in parallel aren’t just 5x faster — they’ve fundamentally rethought how work gets structured. The bottleneck moved from execution to direction. From “can we build this” to “what’s the right workflow, what’s the right spec, what’s the right output.”
The solo founder model works for some categories — high-margin transactional businesses with established playbooks and available third-party regulated infrastructure. It works less well for deep-tech, complex enterprise procurement, physical supply chains, and anywhere the founder cannot realistically be the sole human backstop.
V. So what is actually rare now?
If breadth was never the rare thing — if breadth was mostly translation work that AI now handles — then what is?
Taste. Judgment. The ability to know what to build, why it matters, when to stop, and which AI outputs are actually any good.
The PM who can write SQL is no longer rare. The PM who can look at a dashboard, identify the second-order behaviour buried inside the first-order chart, and connect it to a strategic decision the team isn’t yet asking the right questions about — that PM is rarer than ever.
The designer who can code is no longer rare. The designer who can hold the entire product narrative in their head, push back on a flawed brief, and know when a beautiful component is solving the wrong problem — that designer is rarer than ever.
The founder who can do everything is no longer rare. The founder who can spec the system well enough that AI executes it correctly, recognise their own blind spots, and resist the temptation to scale on hallucinated competence — that founder is, in 2026, the actual unicorn.
The new rare thing isn’t breadth. It’s discernment.
VI. The spec is the product
The way I work now is the inverse of how I hired then.
I don’t write job descriptions that ask for unicorns. I write specs that are precise enough for AI agents to execute and humans to review. The judgment lives in the spec. The execution lives downstream.
The mountain hasn’t moved. The bottleneck did.
The unicorn was never the rare animal. It was the rare animal we conjured up to disguise the structural under-resourcing of our own teams.
AI didn’t create unicorns. It made them unnecessary.
And it left us with the only question that ever mattered:
Do you actually know what you’re building, and why?