There are now very few programmers in the world who aren't using some form of AI co-pilot. But asking technical questions and getting debugging support using GitHub Copilot and Cursor.AI may be just the beginning. AI coding may one day involve agents that can write programs themselves based on natural language prompts. Such programs could even replace human engineers.
AI coding startups that can generate code from natural language prompts include Replit and Bubble.
Some VCs believe that eventually, companies will hire fewer human engineers and each person will manage an AI coding agent. “It's not picture-perfect. It's in the near future, but not today,” said VC Corinne Riley, a partner at Greylock, on stage at TechCrunch Disrupt last week.
He added that many of Greylock's portfolio companies already generally allow coding assistants to participate in coding technical interviews for potential candidates.
But she doesn't think that really young companies should use AI agents in place of human engineers to save cash. At the seed stage, you're saying, “What you're doing is building the foundation of a company, right? So if you're making big engineering trade-offs at that stage, it's probably not the right decision. Those are decisions you can make in the future,” she said.
But money management is also why young engineers at startups should make the most of AI coding assistance, and they should do it now as much as possible, Hustle Fund co-founder and general partner said on stage. Rebutted VC Elizabeth Yin.
“One of the big challenges in the early stages is not knowing exactly what problem you are solving or what the ICP is doing. [ideal customer profile] That's exactly what they need. So you end up giving up a lot of work. So to learn fast, the faster you can go and iterate faster, the better,” Yin said.
She says early-stage startups are looking for anything that allows founders to quickly hack a sample product and get up and running quickly, even if they have to rebuild everything more thoughtfully and carefully later. I believe in embracing the tools. “If it means you can learn that quickly, I’m actually in favor of that,” she said.
This is in contrast to the pre-AI era. Back then, all pilots had to be coded by a skilled person. Now, engineers can prompt models, use AI debugging, and take a peek.
VC Renata Quintini, co-founder of early-stage Renegade Partners, agrees.
“If your goal is to discover or test product-market fit, you should leverage that leverage, but don't worry about optimizing this at the seed stage,” she says. Told above.
Interestingly, we may witness the first germs of the future AI agent workforce as startups founded in 2024 begin operations using AI development processes. And the first people to welcome AI agents as colleagues will be programmers themselves. It's a thought that is both ironic and prophetic.