Give Business Users a Computer, Not a Platform

Airtable’s $1.285B sale marks the end of the low-code abstraction. Give business users an isolated micro VM — a real Linux box — instead of a platform.

The rule I keep coming back to with LLMs is to give them the simplest tool that will do the job. Markdown, not a proprietary format. A filesystem, not a retrieval layer. And if you follow that far enough, a plain Linux box instead of a platform.

That last one is the part I want to argue for.

Airtable sold to Bending Spoons this week for $1.285 billion. The business is healthy, roughly $480M ARR and still growing 20%, so what got marked down wasn’t the revenue. It was the belief that the category gets bigger from here. I don’t think that’s a story about AI writing code. I think the abstraction stopped being the valuable part.

I spent the better part of ten years building low-code platforms, from simple spreadsheet-style builders to large force.com deployments. The pattern never changed. A business user could describe a workflow perfectly and still not get to a working application without someone technical sitting next to them. Salesforce and ServiceNow have enormous partner ecosystems and that isn’t a coincidence. That ecosystem is the product’s missing half.

The partners also steered our roadmap. We kept trying to make the product simpler and they kept asking for more control: record manipulation, database migration, proper error handling. A developer once asked me to add try/catch to our scripting language. We did, along with a lot of other functions no business user was ever going to open.

What we had built was a development environment with tape over the sharp edges, and the tape was there because the user couldn’t program. An LLM removes that constraint and you’re left holding a lot of tape.

Notice where the difficulty went, though. Writing the application takes two minutes now. Everything after it is the problem: somewhere to run, a URL, a login, storage that survives a refresh, mail, a schedule.

That is the case for micro VMs.

A micro VM is a real Linux machine that boots in under a second and is isolated at the hypervisor rather than by a container namespace you are hoping holds. It costs little enough that you can give one to a person, or to a single application, and not think about it. If something goes wrong the blast radius is one box.

Give one to a business user and most of what was missing shows up at once. A filesystem, which is where these models are fastest anyway. Cron. Outbound mail. A package manager. An internal hostname behind the SSO you already run, so what your colleague gets is a link rather than a screen share.

Then there’s the part I find most interesting. Whoever you shared it with can open the box and run their own coding agent inside it. The application arrives with the means to change itself. Nothing in the low-code era worked that way. You filed a ticket with a partner and waited three weeks.

Airtable’s answer was to put AI inside the abstraction. I would spend the effort removing it instead. Give a business user a computer, keep it isolated and secure and reachable inside the company network, and the model will work out the rest.


← All writing