Every Vendor Wants to Be Your AI Operating System. Own Yours Instead.

Within a year, every software vendor you pay will tell you they sell an AI operating system. Most already do. Microsoft, Salesforce, ServiceNow — the category has a name now, buyer's guides, a leaderboard. And the moment a phrase lands on every homepage, it stops telling you anything.

So let me tell you what the vendors mean by it, why their version quietly works against you, and what the version worth having actually looks like — the one built for a mid-sized company that already runs on real systems and can't afford to start over.

What a vendor means by "AI operating system"

Strip the marketing and it's this: a platform you buy that wants to sit at the center of your stack and pull everything else toward it. Agentforce for Salesforce. Copilot for Microsoft. A close agent for your ERP. Each one is real, and each is engineered to make its own product the gravitational center — because that's the business model. The more of your process that lives inside their platform, the harder you are to pry loose.

There's a second problem underneath the first. A lot of what's sold as "agentic" is repackaged — yesterday's chatbot or RPA script with a new label. Analysts already have a name for it: agent washing. Gartner expects 40% of enterprise apps to carry task-specific agents by the end of 2026; today, fewer than one in five companies has deployed an agent that works across teams. The claim is running well ahead of the thing.

The real operating system is your process

Here's the shift. The value of your ERP isn't its code — it's twenty years of accounting logic, tax handling, compliance, and audit trail baked in. No mid-sized company should rebuild that, and nobody should let an AI vibe-code a replacement for it. Keep it.

But the connective tissue between your systems — who does what when, how a quote becomes an order becomes revenue, how an exception gets handled — was never really software. It was people, email, and tribal knowledge. That is the layer AI can finally own. And it's proprietary by nature, because it is your company's process knowledge — not a product anyone can sell off a shelf.

The operating-system metaphor is exact. Your systems of record are the filesystem. MCP — the new standard that lets AI reach into your software directly — is the device interface. The AI layer is the kernel: scheduling the work, enforcing who's allowed to touch what, running your processes as programs. Off-the-shelf substrate. Proprietary orchestration. You don't rip your systems out, and you don't rebuild them. You run them.

Every vendor wants to sell you their AI operating system. The one worth having is the one you own.

Eric Lovgren, lovgren.ai

Why the people selling it can't build it

This is the part that matters, because it's why the gap is still open. The layer I'm describing sits above every vendor and stays neutral to all of them. Look at who's offering to build it for you, and none of them structurally can.

The platform vendors can't build above themselves. Agentforce will never tell you to treat Salesforce as a plain system of record and let an outside layer run the process; Copilot won't make your ERP the source of truth over a Microsoft product. Neutrality is the one thing a lock-in business can't sell — because neutrality is the product, and they're a party to it.

Their consultants can't either. Most are paid through the vendor's partner program, rewarded for expanding that vendor's footprint. A partner comped on Salesforce licenses cannot, in good conscience or good economics, architect a layer whose whole job is to keep Salesforce swappable.

And your own team can't, because the org chart won't let them. The seam between systems is nobody's job by construction: your Salesforce admin stops at the opportunity, your ERP admin starts at the sales order, and the CFO owns the gap in the middle by default. At your size there's usually no platform team and no chief AI officer to hand it to — and hiring one is a $300–650K bet on a person who'd need to be fluent across all your systems and the AI layer at once, and who'd still only ever see your one company.

So who can build it, then?

Almost no one, today — which is exactly why it's worth doing. The work falls between every stool, because it takes four things at once:

1. Neutrality. No stake in which vendor wins — that rules out the platforms and their partners.
2. Real cross-system delivery. Genuine AI-and-MCP capability across your systems, with receipts — not ChatGPT bolted onto a Zapier zap.
3. Someone who operates it. Embedded and running the layer, not a project team that ships and leaves.
4. Willingness to hand you the keys. You own the result — which no business built on lock-in will ever offer.

The person who does all four is a new kind of role: an independent, cross-platform AI applications architect who sits on your side of the table and runs the layer like a fractional executive. That role barely exists yet. I've spent the last two years becoming it.

We run on one

This isn't a theory I'm selling. Our own firm runs on the exact thing I'm describing. Salesforce, our CMS, time tracking, email, scheduling, and our data provider — six systems, orchestrated by a proprietary layer of skills and process definitions that I built and operate. It runs the whole business with one and a half people, and it rebuilt none of the underlying software.

We install what we live in — and the same pattern already runs inside a paying client engagement, proof that it travels beyond us. When something breaks because a vendor changed an API overnight, that's not a flaw in the model; that's the job. An operating system needs an operator. That's the seat.

What this looks like as an engagement

We embed as your fractional AI applications architect and build the layer where your money actually moves: the quote-to-cash spine — Salesforce, your ERP, and the integration between them — with AI working across all of it. Not full-stack development, not infrastructure. The applications that run your business, and the intelligence that runs across them.

You own it. Every engagement lives in your repository, on your credentials, and stays yours when we're done — not mine, not Salesforce's, not Oracle's. That's a promise no platform can make, and it's the whole point.

For the deeper mechanics, two companion pieces go there: when to reach for Salesforce's own AI versus an external one, and the shorthand I use for the whole choice — Claude for thinking, Agentforce for reacting.

Curious what running on one would look like in your business?

The fastest way to find out is to walk your quote-to-cash path out loud for thirty minutes. It's usually enough to see where the glue is costing you — and whether a layer you own could take it over.