There's a place for tech founders who can't write software - but they need technical partners: David Richards
More and more of the software founders I meet at Yorkshire AI Labs cannot write software. It has not held them back.
They describe what they want in plain English to an AI tool such as Lovable or Base44, and a few days later they have a working app, login page and all. There’s even a name for it, vibe coding, and I’m all for it.
I’ve written here before that Mark Zuckerberg might not be the man he is today had these tools existed in 2003, when the Winklevoss twins and a classmate wanted someone to write their campus social network. Today they wouldn’t need him.
But a first version isn’t a company. Friendster had the idea and the users before Facebook and lost them when its pages took up to 40 seconds to load, whereas Zuckerberg waited until his servers could cope before opening to new campuses.


That gap shows the day one of these founders pitches to a bank or a supermarket chain, and the pitch is quickly followed by a lengthy questionnaire and then auditors. The key questions are: where is the data stored? Who has access to it? How would it cope with 10,000 people logging on at once?
The founder has no credible answers, having delegated the engineering to a machine, and nobody in the company knows what it built. AI has democratised building, but it has not democratised engineering.
The founder of Moltbook, a social network for AI agents launched in January, announced that he hadn’t written any of the code himself, and within days the security firm Wiz discovered that its database allowed unauthorised read and write access, exposing 1.5 million access keys.
Meta, Facebook’s owner, bought the company in March. The founder had built something valuable without making it secure. Zuckerberg has engineers.
The machine produces the work. The human takes the credit, often without knowing whether the work is any good. An experiment at Aalto University in Finland gave 246 participants 20 questions from the American law school entrance test to answer with ChatGPT’s help. On average they got 13 right, three more than people answering without AI, yet they believed they had got 16 or 17.
Those with greater knowledge of AI were also more prone to overconfidence. ChatGPT had helped them, but they couldn’t tell how much. That is the illusion of competence, and it isn’t confined to software.
Then there are people demanding far higher pay on a chatbot’s say-so. We have had them.
Figures, the pay data company, put 10,000 salary questions to ChatGPT for jobs in London, Paris and Munich, and six out of ten answers came back too high.
The law is going the same way. In one case a man representing himself at the Employment Appeal Tribunal used ChatGPT to produce a 300-page document of almost 132,000 words, where the guidance is five to 15 pages, and asked the opposing side to list its errors by 4pm the next day. At the hearing he agreed that much of it was not credible. He had left the writing to ChatGPT and the checking to his opponents.
A conductor plays nothing on the night of a performance. Yet you would struggle to find one who can’t read the score, and most spent years playing an instrument before picking up a baton. They know what a horn can do without being able to play one. I wrote the code myself in the first companies I founded, which is why I know what to ask of these machines. A founder who has never written software needs a technical partner early on.
Franz Strauss, principal horn in Munich for 40 years, is said to have warned that an orchestra knows who is master before a new conductor lifts the baton. A bank’s security team takes about as long.
David Richards MBE is managing partner of Yorkshire AI Labs