Bring Your Own AI

For the past year I've noticed a pattern.

Every product eventually wants AI.

Replace forms with conversations. Let users upload a few documents instead of filling out twenty fields. Summarize a website when someone pastes a URL. Generate a course from existing material. Write a news post. Query analytics using natural language.

We've built several of these experiences at Commons.

They work.

In fact, they often work incredibly well.

The problem is that each feature becomes its own product. You're constantly tweaking prompts, switching models, tracking costs, and hitting latency and API limits.

That doesn't scale.

At the same time, I found myself solving another problem.

People wanted to ask questions about data. SQL is far too expressive to expose directly, so we built small domain-specific languages that an LLM could generate safely and deterministically. The model became a translator instead of the source of truth.

That pattern worked so well, it got me thinking why is the AI inside my application at all?

What if the application simply welcomed agents as first-class users?

Instead of embedding an AI experience everywhere, give users secure agent tokens with exactly the same permissions they have. Expose every capability through a clean API. Whether that's MCP or something else almost doesn't matter.

Now users bring their own AI.

The application still has a great interface. But users can also ask Claude, ChatGPT, Gemini (or whatever comes next) to interact with it on their behalf.

Suddenly everything gets simpler.

Your AI already knows how you like to write. It already speaks your language. It already has access to the files you've chosen to share. It can translate content, explain concepts, upload structured information, generate reports, build visualizations, automate repetitive work, or combine data across multiple applications.

And I don't have to guess which model you'll want to use next year.

The application becomes stable.

The intelligence becomes portable.

Maybe this is a bit like what happened with identity.

Websites used to own usernames and passwords. Today it's perfectly normal to arrive with an identity managed somewhere else. The application doesn't care how you authenticated, it only cares what you're allowed to do.

I wonder if AI evolves the same way.

Instead of every product trying to own its own intelligence, perhaps products simply become places where your AI gets work done.

We've started calling this Bring Your Own AI (BYOAI), and we're gradually rolling it out across our apps.

Maybe I'm late to the party.

Maybe everyone will keep embedding chatbots forever.

Or maybe, in a few years, trying to build your own AI into every product will feel as unnecessary as building your own login system.

I don't know.

I just know this direction feels surprisingly liberating.