Saturday, 17 October 2026In AI
Search
Prose
Assistants on Rails

Assistants on Rails

Unless you're a lab or a neolab, there's no business model in offering a personal assistant as a product.

You're wrapping something the labs already offer, or soon will.

Let me be clear: you'd be competing head-on with your own supplier, who already sells to the people you want to sell to and has several OOMs more distribution than you can muster. And you'd have almost no differentiation meaningfully better than what they offer.

My thesis: if you're going to sell personal assistants, build them on a hyperniche, super-valuable platform with deep proprietary rails. Rails the labs can't casually reshape their products to compete with, at least not without bending those products out of shape.

Then add a flywheel: value that compounds into the platform, which in turn powers, and is only accessible to, your agents.

Agents are solved; ride the tide of agentic capabilities the labs ship. Build the backend platforms that unlock highly valuable, specialized niche use cases for your personal assistant agents.

Sell specialized personal assistants bundled with your proprietary platform.

Further reading

What happens when AI companies compete with their customers? | Brookings, brookings.edu https://www.brookings.edu/articles/what-happens-when-ai-companies-compete-with-their-customers/

"Surfing the Frontier Wave that rests between The Bitter Lesson and No Free Lunch", Talaria Scientific https://talariasci.com/blog/no-free-lunch-bitter-lesson

Jamin Ball, "Systems of Record Won the SaaS Era - Clearinghouses Will Win the Agents Era", Clouded Judgement, 2026 https://cloudedjudgement.substack.com/p/systems-of-record-won-the-saas-era

1Solved means the labs own that problem, and will keep improving on it, regardless of your work. Your workarounds will eventually become table stakes.