Starting a technology company used to require a fairly predictable combination of resources, with engineers needed to build the product, capital required to pay them, time needed to get something into the hands of customers and, if things went well, an increasingly large team to turn that initial product into a functioning business.
AI is beginning to change those economics, allowing founders to build products faster, operate with smaller teams, and reach the market with significantly less capital than would previously have been possible. McKinsey research found that generative AI tools can reduce the time required to generate code by 35 to 45 percent, while code documentation can be completed 45 to 50 percent faster.
Anthropic’s analysis of 500,000 coding interactions gives another indication of where this is heading, finding that 79 percent of conversations using its Claude Code product involved automation rather than simply assisting a human developer. Startups also appear to be moving faster than larger companies, with Anthropic estimating that 33 percent of Claude Code conversations related to startup work, compared with 13 percent for enterprise applications.
For entrepreneurs, this looks like unequivocally good news. But I think cheaper, faster execution is available to your competitors too. Building becomes easier, maintaining an advantage becomes harder.Building is becoming the easy part
Startups are also operating with smaller teams. Carta data shows that average headcount at Series B companies fell from 53 employees in 2023 to 45 in 2025, while Series D companies saw average headcount fall 29 percent from its 2023 peak to 131 employees.
There are obviously many forces behind those numbers, and it would be simplistic to attribute the change entirely to AI, but the direction is important because smaller teams can increasingly accomplish things that would previously have required considerably more people, capital and time. Code that once required an engineering team can increasingly be produced by a handful of people working alongside AI, allowing a founder to move from an idea to a prototype faster, test it with customers and iterate without immediately building a large organisation around it.
That should produce more entrepreneurship and experimentation, both of which are good things, but it should also produce far more competition because the same barriers being removed for one founder are being removed for everyone else.
If I can build something in weeks that once took a year, there is a good chance somebody else can too, while if an AI tool allows my team to replicate a competitor’s feature quickly, my competitor has access to exactly the same capability. Speed will continue to matter, particularly for startups competing against slower incumbents, but speed alone becomes much less defensible when everyone is getting faster.
Capital can accelerate the problem
Investors have understandably followed the opportunity, with more than 60 percent of all venture capital raised by companies on Carta in the first quarter of 2026 going to AI companies, the highest proportion it has recorded, while 83 percent of capital invested in SaaS went to AI startups.
There is an enormous amount of genuine innovation contained within those numbers, but more capital and more new companies do not necessarily translate into more durable businesses, particularly when many of those businesses are being built using increasingly accessible technology.
This is where I think founders need to be particularly disciplined, because in a market moving this quickly, raising money, shipping a product or experiencing an initial surge in users can easily be mistaken for evidence that a company has established an advantage. Sometimes it has, but sometimes it simply means the company has arrived early in a market where the barriers to entry are rapidly disappearing and dozens of competitors are likely to follow.
As an investor, the question that interests me is therefore changing from whether a team can build something to understanding why somebody else cannot easily replace it, because those are two very different tests of whether a startup has the foundations to become a lasting company.
What cannot be copied?
That distinction should change what founders obsess over, because while the strongest companies of the AI era will still need to build excellent technology, their defensibility will increasingly come from what sits around that technology.
It might be proprietary data that improves as more customers use the product, a network that becomes more useful as it grows, distribution that competitors struggle to access, a trusted brand in a market where trust matters, or an unusually deep understanding of a customer problem that allows the company to keep moving ahead of everyone trying to copy it.
This is also why I believe founder-market fit becomes more important rather than less important in an AI-driven economy, because when the technical barriers to building fall, insight becomes considerably more valuable. A founder who has spent years inside an industry, understands where customers are frustrated and knows why existing solutions have failed has something considerably harder to replicate than a feature set, particularly when that understanding informs every subsequent product and commercial decision the company makes.
The defining question for the next generation of entrepreneurs will therefore not simply be whether they can build something, because increasingly the answer will be yes, but whether they can build something that becomes progressively more difficult to replace as the company grows.
AI will help more founders launch companies, but as an investor, I am less interested in how quickly they can build than in what will keep customers choosing them once competitors can build the same features. That is a harder question to answer, but it is also a much more useful one when deciding which companies have the potential to last.
