The Risk of Hyper Innovation

The very real risk to tech startups in the age of AI

I recently wrote a paper with a friend (Gemini 3 Pro) about the very real risk to tech startups in 2025 posed by Hyper Innovation.

Here are three examples of why this threat is so real:

1. Frontier LLMs are improving their models at a pace that makes new tech look like old tech in six months.
2. The person sitting at the desk across from you can build your product just as easily as you did (leveraging newer tools).
3. Frontier LLMs are ‘Sherlocking’ app startups.

We are getting ready to launch Escalator, and there is already an upgrade planned for the ‘campaign’ interface because the landscape has changed so much since we started designing the product 10 months ago.

So we’re planning to upgrade a product we haven’t launched yet!

The Threat Has Changed

As LLMs have improved the quality of their output, and people are increasingly using NLP (Natural Language Processing) to engage with technology, we have to be at the forefront of these changes.

If we aren’t, someone else will be.

Most of what you read these days about the rapid advance of AI (specifically software) and the impact on tech startups is the potential for anyone to ‘get rich’ building a business without any technical knowledge.

What they don’t tell you is that the person sitting at the next desk can replicate your product using newer tools.

In ‘my’ thesis the potential threat to startups (depending on complexity and market) has been reduced to weeks from years.

What Is ‘Sherlocking’?

The other thing you don’t hear a lot about is called ‘Sherlocking’.

You may recall a software company in 2000 released an app called Watson that helped Mac users access content faster than the native Apple-designed search function which was called Sherlock. By 2002 Apple released Sherlock 3 which crushed Watson.

The term ‘Sherlocking’ was born.

The end game of LLM’s is to ‘Sherlock’ app startups that leverage their technology.

Make no mistake. If you want to use an app to build a business plan, you can probably get an LLM to produce one for you with some smart prompting at a fraction of the cost.

How Do You Stay Ahead?

There are ways to ‘hold off’ replication of your product or ‘Sherlocking’ like the use of on-premise mini-LLMs trained for a niche business purpose, or the ability to build models that rely on training datasets not available to frontier LLMs.

This is one of the reasons that the strategy of an technology startup has evolved in the world of AI.