Building AI with old software is like trying to make a car out of a horse-drawn carriage, says YC general partner

Business Insider

26-05-2025

AI may be powering a new generation of apps, but most developers are still stuck in the past, said Y Combinator general partner Pete Koomen.
On an episode of the "Y Combinator" podcast published Friday, Koomen likened some AI tools to the earliest versions of automobiles — when inventors simply bolted engines onto wooden carriages without rethinking the design of the vehicle itself. Just like early car builders, developers are taking a revolutionary technology and retrofitting it into legacy designs.
He called this approach the "AI horseless carriage."
"There are all sorts of problems with that design," said Koomen, who is also the founder of software company Optimizely. "Inventing the motor was only a small part of what was needed to produce a vehicle that could take advantage of the enormous power," he added.
The system needs to be redesigned for AI to become useful, he said.
"We're using old software development mentality, techniques to build these features, and we're not actually taking full advantage of what AI can do," Koomen said.
Koomen's comments come as the industry grapples with a quick AI-induced transition for once-hot tech jobs.
Some developers and engineers may be coding themselves out of a job. As AI gets better at writing code, some product managers have speculated that AI will increasingly take on some technical coding tasks and circumvent their need for engineers. Job postings for software engineers on Indeed have hit a five-year low.
At some companies, the engineer's role is transitioning from just writing code to using creative thinking to solve problems, BI's Amanda Hoover wrote in a report in February.
"If a developer is not creative, then you can replace them very easily," said James Stanger, the chief technology evangelist at CompTIA, a nonprofit trade association for the US IT industry.
Gmail's AI is a textbook example of the problem
Koomen pointed to Gmail's AI as a case in point — he called it a chatbot bolted onto an old interface instead of being rebuilt from scratch.
"When the Gmail team set out to build this, they kind of asked, 'How can we slot AI into the Gmail application?" he said.
That approach misses the mark, Koomen said. The AI-generated drafts don't sound like what the user would write, and the prompts required to get the right tone often end up being as long as the email itself.
This makes using AI frustrating and adds more work, he said. Instead, Gmail and other AI tools should give users control over editing the system prompt itself — a set of instructions given to an AI model that guides how it should behave or respond to users.
"By editing this system prompt, I'm able to explain to the AI model how I write emails in general so that I don't have to do it every single time," he added.
The problem, Koomen said, is that developers still treat AI prompts the same way they have treated source code for decades — hidden behind interfaces, tightly controlled, and inaccessible to the user.
"For as long as we've had a software industry, there's been a division of labor between me, the user, and you, the developer," he said.
That model needs to change, Koomen said, adding that this presents a massive opportunity for founders.
"Almost every tool that we've been using for decades can be rethought from the ground up with AI," he said.