Menlo: 76% of enterprise AI was bought, not built, and that's now a design call

The last three years trained you to think about product the old way. Write the spec, make the build-versus-buy case, hand it to a team, ship features. AI knocked the legs out from under all of that at once. Building got cheap. Personas got faster to fake. Solo founders got loud. And the numbers on who gets hired changed. Let me catch you up.

The deep cut

• Making the thing is not knowing if it's good. AI collapsed the cost of building, but Patrick Neeman's point holds: judgment stayed expensive.
• Budget the upkeep or the build bankrupts you. Neeman budgets 30% of build cost yearly, and vibe-coded tools still beat the subscription they replaced.
• A persona with a headshot is a hypothesis. Productside's Rachel looked validated until CRM data broke two of her three claims.

The cheap part got cheaper, the hard part didn't

A designer with a coding agent can stand up an internal tool before a vendor returns the sales call. That's real, and it changes your estimates. But cheap building hides a cost. Neeman budgets 30% of the upfront build cost every year for upkeep, and his math is sobering. Two people for two weeks runs $24,000 up front plus $7,200 a year. A 25-seat subscription runs $6,000 a year. Over three years the build costs $45,600 against $18,000, and only one came with a support team.

The lesson repeats across the sources. Marty Cagan's team at SVPG puts it plainly: the hard part is no longer building, but figuring out what to build. Companies tried shipping fast and seeing what sticks, and got churn and lost trust for it. Cheap production did not buy you better outcomes.

Buy versus build is now a UX call

This used to be a finance decision. A PM wrote the case, procurement signed, designers inherited whatever won. Now the question is where the complexity lands: in your code, in a vendor's roadmap, or on your users. Menlo Ventures found 76% of enterprise AI use cases were purchased rather than built in 2025, up from 53% a year earlier. MIT's NANDA research found purchases from vendors succeeded about 67% of the time against a third of that for internal builds.

Buy and you import another company's org chart and their theory of your user. Build and you own every line forever. A careless mix pushes the cost onto users one copy-paste at a time. Workers in one study switched apps roughly 1,200 times a day, losing about 9% of their week just reorienting. The sharp rule here: buy the commodity, build what you compete on. A vendor can sign the same contract, but can't copy how your suggestion explains itself or how someone corrects it.

The persona that reads true and isn't

You can generate a full persona in 90 seconds now. Name, photo, backstory, a confident claim about who signs the contract. It looks exactly like one built from 30 interviews. Productside tested an AI persona named Rachel against CRM records and support tickets, and two of her central claims fell apart inside fifteen minutes. Her buying-authority claim was wrong. The finance ops manager was the champion, not the buyer.

Two reasons the polish misleads. Stanford research found AI models side with the user roughly 50% more than a person would, so you get a buyer who fits your prompt, not one who exists. And synthetic answers stay shallow. Rank your persona claims by what being wrong costs you. Wrong age costs nothing. Wrong pain costs the roadmap. Wrong buyer costs the deal. Check buying authority, primary pain, and desired outcomes first, using data you already have.

The solo-founder story sells the wrong lesson

There's a loud belief that if you can't build your product alone with AI, you shouldn't build it. Carles Reina, who backed ElevenLabs, called a founder who "literally cannot build products" an instant red flag. More than a third of new startups are now solo-founded. Dario Amodei puts 70 to 80% odds on the first one-person billion-dollar company by 2026.

The confusion is treating the first draft as the finished product. Neeman tells the story of cutting a 40-person project to 8 specialists in 1999, and the point gets twisted today. That move concentrated expertise. The solo-founder pitch removes it and hopes the software fills the hole. Research by Travis Howell and Todd Hall found solo founders hold their own under one condition: deep, broad experience to draw on. The technical founder building for engineers can do it because what they know maps to what the product needs. Knowing a market cold is not the same as knowing how to build something people want.

What the hiring numbers say you should staff for

If you think the whole market shifted to AI, the data says be careful. Aakash Gupta's 2026 report shows PM hires down 45% since 2022, while AI PM hires are up 2.5x. But AI PM is still only about 7 of every 100 PM hires, even as it hit 46% of job postings. The postings are hype. The hires are the signal.

Two moves for your staffing plan. First, don't over-rotate your reqs to "AI PM" when the actual hiring floor is 7%. Second, notice where the demand went: banks and payment networks. JPMorgan Chase, Visa, and Mastercard all grew PM hiring while Meta, Microsoft, and Amazon cut. If you're competing for product talent, you're competing with Capital One now, not just big tech.

Three questions for your team

• For the last internal tool we built, who owns the yearly upkeep, and did we budget 30% for it or hope it maintains itself?
• Which persona claim on our top roadmap bet is riskiest to be wrong about, and have we checked it against CRM and support data this quarter?
• Where are we building a commodity that a vendor sells for $20 a seat, and where are we actually building the thing we compete on?