The Handoffs Were the Quality System: The Rise of the Solo Product Operator
By Mark Cohen
Anthropic's head of growth said on Lenny's podcast in April that when a project needs an engineer for two weeks or less, that engineer now acts as the PM. He described product and design as squeezed, because engineers ship so much faster that the people around them cannot keep pace. Companies are testing a role that does not fit the old org chart: one person who decides what to build, builds it, designs it and tests it.
The industry is starting to call that person "the Solo Product Operator". The definition matters: one accountable owner of the whole loop, working with AI. It does not mean one human with no help. It means one head holding every decision.
The role is no longer rare. Carta tracks tens of thousands of US startups, and the share founded by a single person rose from 23.7% in 2019 to 36.3% in the first half of 2025. Investors still lean toward teams. Of the startups founded in 2024 that went on to raise capital, about 17% had a solo founder. Almost one in five.
"Solo" is generous
The famous examples carry an asterisk. Base44 sold to Wix for $80 million in cash six months after launch, and the headlines called it a solo success. Wix confirmed the company had eight employees. OpenClaw began as one developer's hobby project, went viral, and its creator joined OpenAI. He said that getting agents to everyone needs security research and access to the latest models.
The solo part is the decisions, it seems none of them claim to have had no help.
What a team did that nobody wrote down
Amongst its explicit functions, a team also does a hidden job:
The engineer pushes back on a vague spec.
The tester breaks the thing the engineer called done.
The designer refuses the shortcut.
Each role is a second opinion on the others. While nobody asked for a person to "supply friction" in the job description, the friction caught mistakes anyway. (Side-note: the friction comes from pride, ownership and emotional commitment to one's work. Something I firmly believe adds untold value)
Merge four roles into one person's head and the second opinions vanish, and the data on what happens next is uncomfortable.
METR ran a randomized trial with 16 experienced developers on 246 real tasks in their own large repositories. The developers estimated that AI saved them about 20% of their time. The stopwatch said tasks took 19% longer. These were early-2025 tools, and METR is clear that the result does not prove AI never helps. The part that matters is the 40% gap between belief and measurement. If the person doing the work cannot feel the difference, the person checking the work cannot be the same person.
Veracode tested AI code generation and found that about 44% of tasks produced code with a known vulnerability. Models now write syntactically correct code almost every time. They pass security checks 56% of the time, a rate that has not moved much since Veracode's first report. Veracode sells security tools (including to my current employer), so I suggest reading the number as directional signal only.
DORA's 2025 report, built on nearly 5,000 technology professionals, calls AI an amplifier of whatever an organization already does well or badly. An organization of one amplifies one person's habits, including the bad ones.
Five habits that rebuild the friction
None of these comes from a study. They are purely my thoughts, built from both experience and the evidence above, and they are cheap to follow.
Write the test before the prompt. A written definition of done is the one reviewer that never gets tired or flattered. If you cannot write it, you do not yet know what you are building.
Split the author from the reviewer. Review in a separate session, with different instructions, so the builder never grades its own work. Give the reviewer permission to be rude and direct.
Buy outside eyes where failure is expensive. Authentication, payments and personal data are where fast building breaks. Wiz found a flaw in Base44's platform that let anyone create a verified account on private apps using only a public app ID. Wix fixed it within 24 hours and found no sign of exploitation. A platform that had just sold for $80 million shipped that. A solo operator has less margin. Pay a skilled human to look at login and money.
Keep a kill list. Deciding what not to build is the PM job that no tool performs for you. I run several products and consistently maintain a view of where to invest, where to maintain, where to scale back. Competitor activity, market changes, client preferences and other factors all influence these decisions.
Put real users in early. Your judgment is the least reliable signal in the loop, and the METR numbers say why. Ship small, watch what people do, and let behaviour overrule your opinion.
How to spot one
If you hire for this role, skip the portfolio of finished apps. Ask what the candidate's work got wrong and how they found out. Then ask who reviews it. A good answer names a process, and a weak one names feelings.
In the old world, the handoffs were slow but they delivered value by catching mistakes. The operator's opportunity is to keep the catching and drop the slowness.