Skip to main content

We’re outsourcing to AI

What happens when a business starts outsourcing the busy-work? Is that even useful? Or is it just another cash drain?

Over the last couple of weeks I’ve been working more with AI assistance, and it’s getting hard to insist there’s no “there” there. It’s been able to parse vague error messages from deep in my project dependencies, look up documentation for services I have no desire to become an expert in, and even find logic and security flaws in my code.

Other folks at work who are more deeply involved in building AI services have done some interesting things with it. Work that goes beyond the “let’s vibe-code a to-do list” and gets into problem areas I’m not sure were addressable by either standard coding or human staffing before.

And yet, autocomplete remains a huge mess. It has about a 50% hit-rate, much worse if it’s trying to fill out import statements for me. IntelliSense works just fine here, actually perfectly, but Copilot jumps in the way to suggest packages I don’t have. And when I was setting up comments, Copilot suggested three different syntaxes for querying Astro DB without ever hitting on the correct answer. Apparently hooking up the MCP server was not enough.

There are, of course, many things to worry about with AI. But even if AI doesn’t take our jobs we are effectively moving a lot of that labor out to an external service. That service uses metered billing, is largely out of our control, and often unpredictable. It doesn’t feel like software nearly as much as it does outsourcing.

Do I need to outsource adding import statements to my code? No. But do I get better results that way? Also no. But at least I am now paying for the slower service to do it wrong. That’s much better than my own computer doing it 100% correctly for free.

Outsourcing often seems like a good idea, but it comes at the expense of developing capability. Bert Hubert spoke about this in 2021, although at the time he was talking about outsourcing to other companies, not to AI services. And his example is toaster manufacturing. There’s so much you don’t need to do yourself it’s really easy to end up not doing any of it:

… finally, what is left is the actual toasting element. And I found out, you can just buy these online in bulk. So you also don’t need to make those. And what is left as a toaster manufacturer is that you maintain the logo, you maintain the brand, you write the manuals, you do the logistics of shipping the toaster, and you’re going to be an actual credible toaster company without doing any of the things that are actually around making toast.

What part of the work do you hold onto? And if you get yourself to the point where you’re just the brand, is there any reason someone else can’t come along and take that away from you, too?

Companies that outsource too much lose their ability to innovate.

That’s bad enough, but hear me out here: a big part of the pitch of LLMs is that you can use them to automate the basic stuff, which for some people means “automate the pointless stuff.” It’s of dubious value and even worse news for the environment, but it’s a brilliant business move.

The worst case scenario for people, of course, is that AI takes everyone’s jobs. But even if it doesn’t, what we’re looking at is a situation where a handful of companies have managed to monetize millions and millions of small tasks done in companies all over the world, every day, that used to cost businesses only the salary or wage of the person doing the work.

Every meeting synopsis. Every slide deck generated. That’s a couple of nickels in OpenAI’s pocket, or Microsoft’s, or Anthropic’s.

Because AI is at least human-level unreliable (and often worse), they’ve even monetized minor mistakes. It’s, like, an Austin Powers Dr. Evil plot.

Every so often a business will look around and say “all of this paperwork, all of these meetings, do we really need it?” and maybe some of the waste gets tightened up a bit. But how many AI notes are being made, and billed for, that no one ever looks at? Will we even notice those?

Or, having outsourced the busywork, will we just forget it’s even happening and just go on paying the LLM fees, never knowing just what we are getting. Or even realizing that we’re getting it?

Add a comment
Endmark: No Silver Bullet