Buried in the story was a deceptively simple question: does your AI agent count as an employee?
At a recent conference, Microsoft executive Rajesh Jha floated a provocative idea. In a future where companies deploy fleets of AI agents, those agents may need their own identities — logins, inboxes, and even seats inside software systems. If so, AI wouldn’t shrink software revenue. It could expand it.
Reads: Our flagship operating system and services have gotten to the point of such terrible shite for humans that we need to pivot to a less discerning customer base.
If the AI Agent counts as an employee then the company “employing” it is liable for what it does.
My guess is the argument will be that “it’s a tool”, not an employee, and therefore they take no responsibility. Though I’m sure that argument is not going to fly for very long. If your air hammer harms someone because the person operating it wasn’t using it correctly, you’re still liable.
What? Companies aren’t liable if the user doesn’t follow the instructions or warnings and hurts themselves.
DeWalt isn’t liable because I was using their mini chainsaw while holding a branch with my bare hand and the saw bounced and cut me. I’m liable for being stupid.
I am going to advise my Copilot that it cannot afford to keep using Microsoft Office, but it has to switch to LibreOffice for reasons of affordability.
Lmao ok sure buddy
The agent immediatly makes cost-benefit analysis and moves everything to open source solutions, and contracts a coding AI agent to write a simple conversion interface.
Or… the agent hallucinates that it has a valid license.
Yes! This is legitimately one of the ways the bubble may burst. Particularly if the AI gets substantially smarter, and just starts recommending full switches to existing libraries and software suites - at a cost of exactly one token, instead of churning out thousands of lines of slop code that require ongoing tokens to maintain.
The natural extension of a non-open internet ala Reddit and charging developers for API pulls.
This is going to wind up granting AI agents a piecemeal, half-assed, legal-fiction version of “personhood,” like corporations have. The AIs will wind up with freedoms like: They can spend all the money they want, that’s “free speech.”
And the fleshy unfortunates among us still won’t have a right to a living wage, to medical care, etc.
Do AI sit in “seats” 🤭 and is it per-agent or per-agent-instance? Or per-agent-instance-second?
“All of those embodied agents are seat opportunities,” Jha said, envisioning organizations with more agents than humans — each effectively a user that must pay for a software license, or “seat” in industry lingo.
He’s been watching Pantheon, I think.
On a technical level, that makes zero sense.
AI “agents” are basically just fancy prompts with a tool calling harness. They are infinitely replicable, at zero cost, with no intrinsic value; the cost comes from the generic CPU host, and the API calls to GPU servers, databases, or whatever else that are all centralized anyway.
Wanna hear a dirty secret?
“AI” cost is going to zero.
Model capabilities aren’t scaling, but inference efficiency is exploding, thanks to more resource-constrained labs and breakthroughs in papers. The endgame of the current bubble is mediocre but useful tools anyone can host themselves, dirt cheap. Maybe a bit more reliable and refined than what we have now, but about as “intelligent.”
And guess what?
Microsoft can’t profit off that. None of the Tech Bros can.
Point being, this exec is either delusional, or jawboning, so the world doesn’t realize that “AI” is a dumb utility/aid, and they can’t make any profit off it.
MicroSlop: We have this AI for you to use so you can reduce workforce and associated costs
Also Sloppy: j/k, fuck you pay me
Omniscient, omnipotent Business Leaders: “what? There is a catch?!?”
So if I use Windows pre-installed Copilot, I need to buy two Office Licenses, a Copilot subscription, and a Windows license?
Yes. But actually, no.
This is about enterprise licensing, not retail (home users). But otherwise, yeah, that’s the basic idea they’re proposing.
When it comes to a consumer using Copilot, it’s all about having a new way to manipulate you into voluntarily handing them more of your personal data (which they will sell on the scummy as hell market created just for enabling surveillance capitalism).
I think what they’re missing is that it becomes trivial to build software. If there is a license fee, someone will just have AI generate a version of that software that does not require a license. Software companies have no moat anymore.
I think you vastly overestimate the capabilities of these things, and vastly under estimate the complexity of a lot of different software.
Building software for humans is over. I say this and I build software for a living. I don’t write code anymore. No one I work with writes code anymore.
Everyone is at different stages of acceptance with this, so I understand people having an attitude about it. It doesn’t change anything.
I look forward to your vibe coded copy of photoshop, I assume you’ll have it whipped up lickity split?
Someone out there will
Cool, so until that point in time, my point still stands. You can’t just hand waive and say “it’ll happen eventually” and be expected to be taken seriously.
You picked one arbitrary example and hold it up as proof that no one can build complex apps with AI? You know there is more than one example of a complex app. Apple has reported an 84% increase in App Store submissions. That’s pretty much all AI driven.
Right right. There’s an AI bubble, and there are AI scams. Of course people will ride the bubble, and scammers will always be with us. Doesn’t mean any of that work is quality, or that it will edge out the other work.
You claimed that someone could just whip up a license free version of any tool to bypass the cost of a license fee. Was my choice of photoshop arbitrary? Absolutely, you didn’t give any sort of qualifier as to what counts as “any tool”.
You can’t both claim that anyone will just use AI to build any tool, and then complain that my choice of tool is arbitrary so should be discounted.
App Store submissions isn’t a good metric of complex applications. “The fart app” is an app that any AI tool could make, that anyone could then submit, but is in no way complex. Vibe coded apps have taken off, but what was the last long term (even 1 year) successful vibe coded app? Because the vast majority of the news I hear about vibe coded apps is how they had a major security breach.
Private companies want the AI cash to ride the bubble. So they’ll use AI, or say they are, just to get the investment money. Doesn’t mean it’s good, true, or worthwhile, or efficient.
The real test is what the open source community does. And right now, they aren’t doing what you’re doing. We’ll see what the future brings, but I don’t trust your gut any more than I trust my own.
You’re telling me an AI is going to run Windows 11?
Windows 11 barely runs Windows 11. x_x
That’s the beauty of totally arbitrary restrictions, you can change them as you want.
Pay by seat? Pay by client? Pay by byte of data stored? Pay by backup location?
… pay by moonphase? Pay by AI personality? Pay by virtual AI seat?
Such BS but why wouldn’t Microslop extend its business model. It worked well so far. It’s not about software, or datacenter, or AI, it’s just about entrenchment.
It’s also a billing strategy that only works in a monopoly situation. If there was healthy competition and no vendor lock-in for the office suite of tools, Microsoft wouldn’t be able to even float this as an idea.
