Aug 31, 2023 2 min read

AI copilot for X is the new Uber for X

This week I spent some time looking through the summer 2023 batch of YC companies. What stood out to me is the prevalence of the "AI copilot for X" pitch. It reminds me of the "Uber for X" pitches that were frequently used in the last decade to describe marketplaces.

Copilot companies in the S23 batch

I counted 14 companies out of 216 that used some variation of "copilot" in their product pitch. Another 2 didn't use this phrase, but were essentially building copilot products. 16 companies, or ~7% of the batch, is a lot!

I define "AI copilot" products as having the following characteristics:

There is a human in the loop. So, it's distinct from AI replacing a job altogether, though there are a few of these companies in the batch too.
The product utilizes generative AI to help someone with their work output. For example, it helps someone move from a writer to an editor/reviewer. Or, it increases the speed at which content is produced by turning intention into a defined end product faster than a person's brain and fingers can.
The business model is similar to a vertical SaaS. They target operating workflows of specific industries. The value add comes from (1) understanding what the intermediary and end products should look like and (2) hooking into upstream or downstream workflows.

An example of an interesting copilot product from the S23 batch is Agentive, which targets the auditing industry. It's a business idea I've listed in a notepad somewhere. Auditing produces a lot of paper trail. While auditors audit companies they also get audited by regulators. Producing this paper trail is not revenue generating, takes up a lot of time from junior auditors, and is not particularly gratifying. It's the perfect space for copilots to step in and help!

Startups that enable copilot companies

I also counted ~6 startups building tooling that enable the "AI copilot for X" companies. These tooling startups are focused on the technical systems that will make copilot products compliant, scalable, and easier to build. Tooling that software engineers rely on for typical SaaS businesses need to be reconsidered for generative AI. Think security, authentication, testing, and infrastructure.

An example of an interesting tooling product from the YC batch is Talc. They are building a staging environment for LLMs. One of the challenges of building copilot products is that there is so much variability in what a user can input in their prompts and what an AI will return (think AI hallucinations). This makes it hard to guarantee high quality customer experiences. Talc makes it possible for product teams to test a variety of prompt scenarios in a scalable way. That's super useful.

Talc's investment pitch could easily be "BrowserStack for AI Copilots." And, I expect many tooling companies to use the "X for AI Copilots" tagline.

Predictions

There will be two types of companies that survive as standalone businesses:

"Copilot for X" companies where (1) the incumbent SaaS products in the vertical are too slow to incorporate AI and/or (2) the workflows for the vertical are still manual and rely on copy/paste with MS Word or other undifferentiated software.
"X for AI Copilots" companies that build tooling that exists for SaaS application development, but applied to the AI development lifecycle. It's more difficult for existing developer tooling companies to pivot to serving AI products.

The biggest threat to most "Copilot for X" startups is that incumbent SaaS companies can bring more value to customers by introducing copilot into existing products/workflows than a standalone copilot can. The question becomes: can a startup integrate with the rest of the industry's workflows faster than an incumbent can build a copilot into their product? I predict there will be a lot of acquihires of talented startup teams who can't outpace incumbents, especially if the tooling startups mature quickly.