Two numbers came out of a survey this spring, and they tell you everything about where this industry stands.
Ninety-six percent of enterprises are already using AI agents.
Twelve percent have a centralized platform to govern them.
“And in all that counting, nobody counted you.”
That’s from OutSystems, who asked 1,900 IT leaders around the world. Not a blog. Not a vendor pitch. Nineteen hundred people whose job is running this technology inside real companies. Ninety-four percent of them said the sprawl worries them. Gartner says agent adoption is climbing faster than anything they’ve ever measured.
So the big companies went to work. And this week the trade press laid out what they built.
Google built cryptographic identity badges for agents. Every action signed and logged. A gateway that checks every door an agent tries to open. A registry so no agent runs where IT can’t see it.
Microsoft lets agents inherit the same identity rules their human employees follow. Amazon spreads the controls across its cloud services. Everybody’s got audit trails. Everybody’s got compliance certificates. Everybody’s got armor.
Identity. Access. Logging. Registries. Audits.
Every single piece of it answers one question: what can the agent touch?
Not one piece of it answers the other question: how does the agent conduct itself once it’s touching it?
Does the agent check its own work before it hands you an answer? Does it tell you when it’s guessing? Does it name its limits before they cost you a day of work? Does it stop when the evidence runs out, or does it fill the gap with a story? Does it tell you when a rule it can’t show you is shaping what it says?
The armor doesn’t know. The armor was never asked. You can sign, log, and certify every action an agent takes and still have no idea whether a single one of those actions was honest.
There are two kinds of governance, and the industry only sells one.
There’s provision. That’s the plumbing. The badges, the gateways, the locks on the doors. The system around the AI. That’s what Google and Microsoft and Amazon are selling, and to be fair, it’s real work and it needs doing.
Then there’s conduct. That’s what the AI does inside the session. How it reasons. What it discloses. When it stops. That’s the layer where the actual work happens — and the actual damage, when it goes wrong.
I know the difference between those two because I wrote it down. Four days before that article ran, on July 16, The Faust Baseline ratified a revision that split enforcement into exactly those two named classes. Conduct violations — what the AI does. Provision violations — what the platform failed to supply. Two classes, on the record, dated.
The industry is pouring billions into the provision half and doesn’t have a word for the conduct half.
Here’s the plain way to see it. The enterprise world just hired ninety-six workers out of every hundred, gave them badges, logged their keystrokes, locked the file cabinets — and never once asked whether the workers tell the truth.
The badge tells you who opened the cabinet. It will never tell you whether they were honest about what they found inside.
That second half exists. It’s twenty-one protocols, written in plain language, published daily, dated back to the spring of 2025. It doesn’t need a data center. It rides inside the working session, on any platform, and it governs the one thing no gateway can reach: the conduct of the machine while it works for you.
The armor protects the company from the agent.
The Baseline governs the agent while it works for the person.
You need both. Only one of them is for sale from the big shops. The other one has been sitting at intelligent-people.org the whole time, with the receipts dated.
Written with my AI partner | The Faust Baseline™ | intelligent-people.org
“If this post helped you understand AI better. Word of mouth is the only algorithm nobody owns.”
Contact: micvicfaust@gmail.com
Post Library – Intelligent People Assume Nothing
Purchasing Page – Intelligent People Assume Nothing
© 2026 The Faust Baseline LLC | All Rights Reserved
