I Almost Sneered at the AI Gold Rush. Then I Remembered I'm In It.

A good friend of mine hired another shop to set up an Agent. He paid them nearly $6,000 for a semi-dangerous setup.

When he told me, my first reaction was a complicated set of feelings: A little competitive. A little protective of him. And, when he described the setup, a real wince.

Here's what he bought. A small, new outfit — nice website with a big list of "partners", clean deck, the works — walked him through an open-source installer, added a skill or two, and stood the whole thing up on a Mac Mini in his office. Local. Unmanaged. Not backed up. Hooked to his primary email. Running on his own tokens.

And he's part of a wave. There's another guy I know — built a nice t-shirt printing business, caught the AI moment, became a speaker, put out a book, and now he sells agents too. No technology background, no AI background. Just hustle and timing.

My instinct was to be annoyed. And then I caught myself.

Because here's the thing I don't love admitting: I run an AI company that also didn't exist a year ago. I don't like the label "AI agency" — it feels small next to what we're actually trying to build — but if I'm honest, from the outside, that is exactly what we look like. Who am I to sneer? This is a gold rush. There's room for a lot of people to do well in it. And nothing says the one who wins has to be me.

So I let the annoyance go and got curious instead.

Then something rare happened. I got a chance to see what one of these shops actually delivered — not the pitch, the output. I expected to come away rattled. Instead I came away more confident than ever in what we're building.

Not because their work was especially bad. Because the gap between us sat exactly where I'd hoped it would — and it was wider than I expected.

Let me be precise about where, because this is the useful part.

The tools got good. A motivated person with no background can get an agent running in a demo now, and pretending otherwise is its own kind of dishonesty. That part is real.

But the demo was never the hard part. The hard part is everything after: making it run in production, safely, for someone whose business depends on it, for years. The security model. The oversight layer. The thing that breaks at 2 a.m. The data you can't afford to leak. When the easy part got easy, that gap didn't shrink — it became the whole game. And what I saw delivered stopped at the demo.

Is our own work a finished masterpiece? Of course not. We're a work in progress too — building in the open means you get to watch me say that out loud. But there's one line I care about that isn't a matter of taste, and it's this: the work shouldn't leave the customer exposed. Wiring an autonomous agent into someone's primary email and file storage with no oversight or controls isn't a different philosophy of deployment. It's an unlocked door (even with email send disabled for the agent!). I've written on this blog about how the head of AI safety at Meta had her own inbox wiped by exactly this kind of setup. That's the part I can't shrug off — not because of who's selling it, but because of who gets hurt if it goes sideways.

So that's the confidence. Earned by comparison, held with humility.

But the same look in the mirror showed me something less flattering, and this is the part I really want to sit with.

These shops have clients. Real ones, plenty of them. And it isn't only because buyers can't tell deep engineering from a good demo. It's that they're meeting a need I had quietly decided not to meet.

Some of these customers don't want a vendor. They want to learn how to do it themselves.

I'd been filing those people under "not my customer." Too small, too DIY, not the seasoned operator I picture when I describe who we're for. The people my friend hired didn't draw that line. They looked at the same person and saw a customer worth serving.

My friend also saw that this is important new technology that he needs to understand. That is relevant and makes sense - it's at the heart of our mission (Soveriegn AI, Realized), but I missed it.

That stung — because my whole reason for doing this is to arm the independent economy, to put this leverage in the hands of people who didn't get to go to the Big Tech party. And there I was, walking past a roomful of exactly those people because they wanted to be taught, not sold to.

So I did something about it. One week, heads down with the team (actually, some dictated notes sent to Beetle, who tasked the team and monitored progress, but we'll get to that next week!).

We launched self-serve Force Multiplier. The thing my friend paid nearly six grand to have installed once — unmanaged, on a box in his office — you can now run yourself, managed and backed up, with the oversight baked in, starting at $149 a month, tokens included. Higher tiers at $349 and $549 add more support, a bigger token budget, and better models. It's at selfserve.getforcemultiplier.ai.

It's not the same thing those shops sell. It's the version where you actually learn how it works, and your data isn't sitting out in the open. It's self-serve, so customers can still shoot themselves in the foot, but it's default safe and reliable, with support and resources to let them get good at this as well. It's a tool that you can put into your business and rely on, not an expensive learning toy.

Two things I'm carrying out of this.

The first is for you, if you're the one buying. In a gold rush, looking credible gets cheap — anyone can have Claude build a slick website and say they are "partnered" with Anthropic, OpenAI, et al. The signal that still costs something is whether the person across the table can tell you what happens after the demo. Ask about security. Ask what breaks. Ask who owns the risk when it does. Ask what you're paying for - $6k turned out to be a higher hourly rate than my partner-level attornies (for a junior team member who learned about Hermes a few weeks ago!). These are the questions that separate the people who can get it running from the people who'll still be standing behind it in two years.

The second is for me. If you're willing to look hard at what you're actually doing — not the flattering story you tell yourself about it — you'll sometimes find you've been ignoring people you claim to serve. I got lucky. A couple of enterprising people held up a mirror without meaning to, and I didn't love everything in it. The honest move wasn't to be annoyed at them. It was to go build the thing the market was saying was needed — and to get even more serious about the part they can't touch.

Which is the engineering. And the best proof I have that ours is real isn't a feeling — it's that we run our own company on it. Our delivery org is staffed with the same AI employees we sell. That's the next post, and it's the one I'm most looking forward to writing.

What's something you stopped offering because you'd decided it "wasn't your customer"? I'd put money on there being a version of my mirror somewhere in your business too.

— Matt