How AI Has Changed Startups Forever

When ChatGPT first came out three years ago, I wrote a wrapper to help myself prioritize emails and messages. My coding skills were rusty, so I used ChatGPT itself as a coding copilot. I instantly felt the magic. It took me maybe half a day to finish the project, while it would normally take me a few days in the pre-AI era.

During the subsequent couple of years, technical founders that we work with at Alliance started adopting various AI coding tools such as Cursor. I asked them how much productivity boost they have experienced. On average it was about 1.5x. A few said as high as 2x.

Earlier this week, I asked the same question to the most recent Alliance cohort. Their answers ranged between 2x and 4x. I was shocked. Not only by the magnitude of improvement, but also by the speed of improvement — a sheer 4x productivity increase within the span of three years and there is no sign of slowing down any time soon.

When I first used ChatGPT as a copilot, I had a hunch that AI will change the startup game forever. Now, I am convinced of it.

The cost of writing software is trending to zero. But this is not all. Marketing, sales, customer service, operations, and other business functions will experience significant cost reductions too. This is already happening at big companies. Pre-PMF startups are less affected because founders have to take on user acquisition and support themselves in order to stay close to customers. But Post-PMF startups are starting to adopt AI to scale faster.

The corollary of all this is self-evident. Startups don't need a lot of funding and a lot of people anymore to become big. To put it another way, expect more unicorns built by just a couple of cofounders and half a million dollars in funding. We are already seeing more and more Alliance startups in the last couple of years that hit 7, 8, or even 9 figures in annual run rate after raising just a small pre-seed from us and some angels. They are still fewer than 10-20 people today, and will never have to raise again.

This leads to the question: if it's so much cheaper now to write software and to scale businesses, what moats are left?

• Proprietary Data: Exclusive customer data or hard-to-acquire public datasets create defensibility.
• Network Effects: Theoretically you can recreate the software stack of Facebook or Nasdaq, you can't acquire their customers easily.
• Deep Integration: Enterprise software with deep workflow integrations have high switching costs.
• Ecosystem Lock-in: Platforms with strong developer ecosystems and third-party integrations remain sticky.
• Regulatory Barriers: AI can't get you the licenses and the relationships necessary to operate highly regulated industries like finance, healthcare, and defense.
• Trust and Brand: Consumers and organizations will continue to trust established brands for critical decisions.
• Physical World: Physical infrastructure, robotics, and supply chains still have constraints that AI can't bypass.
• Deep User Insights: While AI commoditizes half-decent ideas, real understanding of users is simply not embedded in the training set — it's uncovered through talking directly to users.

The highest priority for pre-PMF startups is to find PMF, but it may be more important than ever to think ahead about potential post-PMF defensibility.