Why Am I Middleware?

This feels like a temporary job.

The other day I found myself copying a response from one AI and pasting it into another AI.

Then copying that AI’s response and pasting it back into the first one.

Then back again.

At some point I realized what I was doing.

I was middleware.

Two remarkably capable artificial intelligences were having a conversation, and apparently the communications protocol between them was me.

This seemed inefficient.

It started with my phone bill

I was considering switching from Verizon to T-Mobile.

Not exactly civilization-changing stuff, but there were enough variables to make it complicated. I still owed money on my iPhone. T-Mobile had an offer to reimburse some of that if I switched. There were different plans, different promotions, an Apple Watch sharing my number, AutoPay discounts, trade-in offers, eligibility periods, and various rules about which things could be combined.

So I started asking AI to help me sort it out.

And it was remarkably good at it.

I wasn’t just asking, “Which plan is cheaper?” I was giving it four or five things to figure out at once. Compare these plans. Include the watch. Account for what I still owe Verizon. Figure out whether this $800 switching offer can coexist with that phone promotion. Tell me what happens if I wait 90 days. And while you’re at it, tell me if there’s something I haven’t thought to ask.

The answers came back almost immediately.

That part is easy to overlook.

Give the same problem to a human representative and they would have to understand the whole thing, look up several policies, keep all the dependencies straight, probably put me on hold at least once, and then explain it back to me.

I’m not even sure I could have gotten the same level of analysis from a human, no matter how long I was willing to stay on the phone.

But there was another problem.

One AI knew what Verizon was doing. Another knew what T-Mobile was offering. And neither could talk to the other.

So I copied.

And pasted.

And copied.

And pasted.

It was like having two brilliant attorneys negotiating a contract while I stood between their offices carrying handwritten notes.

Except the attorneys could think in milliseconds.

And I was using Command-C.

Why am I in the loop?

Once I noticed it, I started wondering why I was involved at all.

My AI already knew what I wanted. It knew what I was paying, what devices I had, what mattered to me, and what questions needed to be answered.

T-Mobile’s systems presumably knew what T-Mobile could offer.

Why couldn’t they just work it out?

My AI could say: Mateo wants to keep his phone and number, keep his Apple Watch working, preserve or improve his service, eliminate as much of his remaining device balance as possible, and lower his monthly bill. Here are the constraints. What can you do?

T-Mobile’s AI could respond.

Mine could challenge something.

Theirs could counter.

Maybe Verizon’s AI gets involved and decides it would rather keep me.

Fine. Let them fight.

Wake me when somebody wins.

Instead, we’ve built systems capable of reasoning through the entire transaction and left the human responsible for moving the text between them.

That feels temporary.

Then I started noticing how much middleware I actually do

Phone service is probably the least interesting example.

Why am I comparing insurance policies?

Why am I watching for subscription increases, remembering to cancel free trials, negotiating internet bills, finding flights, filling out the same information on forms, scheduling appointments, submitting warranty claims, disputing charges, and tracking packages?

None of these things is particularly difficult.

But collectively, we’ve turned every functioning adult into the unpaid operations manager of a small organization consisting of themselves.

A surprising amount of modern life is administration.

And we’ve spent decades trying to make that administration more efficient.

We went from offices to phone calls, from phone calls to websites, from websites to apps, and now from apps to AI.

But I’m still here.

Clicking things.

Remembering things.

Moving information from one system to another.

Being middleware.

Presumably this goes away

The obvious next step isn’t a better chatbot.

It’s giving the chatbot permission to do things.

And we’re already starting to see the next step. Grok Bot, for example, gives an AI its own computer and lets it work across applications, completing jobs end to end instead of waiting for a human to shuttle information between them. In other words, somebody at xAI apparently noticed my new career as middleware and is already working on eliminating the position.

If my AI knows that I want the least expensive cellular plan that meets a certain set of requirements, why should I shop for it?

Go shop.

Negotiate.

Come back when you have something worth showing me.

Maybe it says:

T-Mobile will save you $37 a month. They’ll reimburse $800 of what you owe Verizon. Your watch will still work. You’ll lose one feature you don’t use and gain international data. Switching will cost you $214 after everything settles.

Do you want me to do it?

Sure.

Eventually, even that question might be unnecessary.

I could give it rules.

You may change recurring services without asking me if you save at least 10 percent, don’t reduce the quality of the service, don’t create a commitment longer than a year, and don’t spend more than $500 in the process.

Now my phone plan changes while I’m out riding my bike.

That seems better.

Although there may be a problem with where this leads

I’ve spent a large part of my career making things more efficient.

Fewer steps. Less labor. Less waiting. Less duplication. Fewer opportunities for somebody to screw something up.

That’s usually considered progress.

But if you keep following that logic, eventually you arrive at a strange place.

What happens when we’ve streamlined almost everything?

Not just phone plans and insurance forms. Work itself.

I’ve written about the larger version of that question before, in When Nothing Needs to Be Done: what happens if intelligence and labor become abundant enough that much of the work required to sustain our lives simply isn’t necessary anymore?

This is a much smaller version of the same question.

Maybe we don’t wake up one morning and discover that the robots have taken all the jobs.

Maybe things just start disappearing.

First I stop calling the phone company.

Then I stop comparing insurance.

Then I stop booking flights.

Then I stop paying bills, filling out forms, scheduling appointments, renewing things, canceling things, negotiating things, and keeping track of all the little administrative debris that seems to accumulate around being alive.

Good.

I don’t think most of us will mourn the loss of hold music.

But then what?

For thousands of years, we’ve spent an enormous percentage of our lives doing things because they needed to be done.

For many people, nearly all of life is still organized around survival.

If we actually succeed at removing more and more of that necessity, maybe the question stops being how to make everything more efficient.

Maybe it becomes what all that efficiency was for.

Art. Travel. Learning things for no reason. Building things because they’re interesting. Spending time with people we like. Sitting by the ocean.

Doing absolutely nothing.

We’ve spent an extraordinary amount of effort trying to save time.

It would be funny if the thing we’re least prepared for is actually having it.

But we’re not there yet.

For now, the machines are apparently waiting for me.

Command-C. Command-V.