By now, there’s hardly a coder in the world who isn’t using an AI copilot in some way. But using GitHub Copilot or Cursor.AI to ask technical questions and get debugging help could be just the beginning. AI coding may one day involve agents that can write the programs themselves based on a natural language prompt. Such programs may even replace human engineers.
AI coding startups that can generate code from a natural language prompt include Replit and Bubble, among many others.
Eventually, some VCs believe, companies will hire fewer human engineers and have each human manage AI coding agents. “It’s not pie in the sky. It’s in the near future, but it’s not today,” VC Corinne Riley, partner at Greylock, said onstage at TechCrunch Disrupt last week.
Coding assistants are already commonly allowed during the coding technical interview for prospective hires at many of Greylock’s portfolio companies, she added.
However, she doesn’t believe that AI agents should ever be used to replace human engineers in really young companies, to save cash. At the seed stage, “what you’re doing is you’re building the foundations of the company, right? And so if you’re making major engineering trade-offs at that stage, it’s probably not the right decision. Those are decisions that you can make in the future,” she said.
Yet, cash management is also the very reason why young startup engineers should be using AI coding assistance as much and as well as they can right now, countered VC Elizabeth Yin, co-founder and general partner of Hustle Fund, onstage.
“One of the major challenges of the early stages, you don’t quite know exactly what problem you’re solving and quite exactly what the ICP [ideal customer profile] is and quite exactly what they need. So you’re going to end up throwing out a lot of work. So the faster you can go, and the faster you can iterate, the better in order to learn quickly,” Yin said.
She believes early-stage startups should be open to any tool that lets founders quickly hack together sample products to move faster, even if it would all have to be carefully rebuilt more thoughtfully later. “I would actually be a proponent of that if it means you can learn that much faster,” she said.
That’s in contrast to the days before AI, when every pilot would have to be coded by someone who had the skills. Today, an engineer can prompt a model, use some AI debugging, and take a peek.
In agreement is VC Renata Quintini, co-founder of early-stage Renegade Partners.
“If it’s about discovering product-market fit or testing out, you should use that leverage, but I wouldn’t worry about optimizing this at the seed stage,” she said onstage.
Interestingly, as startups founded in 2024 launch out of the gate using AI development processes, we could be witnessing the seeds of the first future AI agent workforce. And the first people to get AI agents as co-workers would be coders themselves. It’s a thought that is equal parts ironic and prophetic.
