Late in my last corporate role, I sat through a day of vendor briefings, and by the end of it I could not have told you what was wrong with a single one of them. Every vendor was right. The identity firm was right that nobody could say which agent had taken an action. The data firm was right that agents were reading things nobody had cleared them to read. The security firm was right about where the agents ran and what they could reach from there, and the application firms were right that the agents were doing real work, fast, and that the work was worth paying for. Each briefing had a diagram, and each diagram was the whole problem as seen from one chair. When the last of them left I had a stack of true diagrams and no picture.
If you are the leader who has to decide about artificial intelligence (AI), you have had that day, or you are about to. A dozen vendors each describe a different slice of the problem: who an agent is, what data it touches, where it runs, what it does. Each is right about its slice, and none says how the slices add up. The regulations you answer to arrive as further, separate lists. You are left to assemble the picture alone, and a picture assembled alone cannot be defended end to end.
The noise is not the vendors' fault. Each of them has built a good business on one slice, and a good business on one slice has no reason to draw the whole. Nobody owns the connective layer that makes the slices add up to something you could stand behind in front of a board, a regulator or a customer, so the connective layer is left to you, on a whiteboard, after the briefings, with the slides stacked on the chair beside you. I did that for years. The whiteboard was never wrong, exactly. It was never defensible either, because nobody but me had drawn it and nobody but me could read it.
The whiteboard had a second cost, and I saw it only later. A picture that lives in one person's head, or inside one vendor's product, leaves when they do. Each vendor carries its own diagram of your business, and the day a vendor goes, the part of the picture it held goes with it, and you start that part over.
What I wanted was one picture I could put every briefing on. I did not want another product, because I had enough products, or another list, because the lists were the problem. I wanted a map: a fixed set of places, so that when the next vendor walked in I could say, before the second slide, which part of the whole they were describing and which parts they were not. And I wanted the map to be mine, so that it stayed when they left.
The map exists now, and I spent a book drawing it. The Trusted Living Agent Framework is an architecture for agents that carry their own governance, and as a map it is one structure onto which every vendor, product, agent and regulation that creates noise for a leader can be placed and read. The shortest way I know to say it is five pillars by three signatures.
The five pillars are the surfaces of an agent, and together they cover everything an agent is and does. Identity answers who you are and what you may do. Data answers what you know and what you touch. Infrastructure answers where you run. Action answers what you do, and Orchestration answers how you work with other agents. I have tried for a long time to find a sixth, and to find one of the five that another could absorb, and I cannot. An agent is one thing, and all five surfaces are present whether or not anyone is looking at them.
The three signatures are what make a surface living rather than inert. A surface is self-describing when it states what it is, what it is for and what it will not do, on request. It is self-reporting when it reports its own state and activity, so that every action carries its reason and nothing is silent. It is self-governing when it enforces its rules at the point of decision, within limits a human sets and checks, in the moment rather than in a post-mortem the following week. Lay the three across the five and you have fifteen cells.
Beneath the grid sits the business itself, and it matters more than its size on the page suggests. Before anything is placed on the map, the business is on it: who you are, what you are there to do, and what good looks like, captured, so that every agent's role traces upward to a goal the company actually set. Governance that starts anywhere else is governing a workforce nobody has described.
A map earns its name when something unfamiliar can be placed on it, so let me place three things. Take the identity vendor from my day of briefings. They land on the Identity pillar, and when I listen to the slide again with the map in front of me I can hear that they are strong in the first column, describing who an agent is, and silent in the third, because nothing in their product stops an agent from acting beyond its role in the moment it tries. That is not a criticism. It is a placement, and a placement tells me what I still have to buy or build. Take a managed security service: it lands on Infrastructure, mostly in the reporting column. And take the agent in your sales department that drafts the follow-up before the call is over. An agent is a column of its own, because it occupies all five rows at once, and the honest question is which of its cells it has reached and which are empty.
Placing is the whole trick. Once you can place a thing, you can see what is beside it and what is missing, and the stack of true diagrams becomes one diagram with many labels on it.
Placing also changes what a vendor is to you. A vendor on your map holds some of its cells and does not own the picture. When a vendor leaves, you can see which cells go dark and what has to fill them, and everything else on the map stays where it was. That is the difference between changing a supplier and starting over.
Each surface also matures in order. It describes itself first, then reports itself, then governs itself, and the order does not run backward: an agent cannot refuse an improper use of data if it cannot say what data it holds. That order lets you stand in front of any cell and ask one plain question, which of the three it has reached, and the answer is a word rather than a number.
I spent a career on the perimeter, so I want to be plain about what the map does to the governance you already run. Nothing on it asks you to take a fence down, and nothing on it stands in for a policy, an audit or a review board. The perimeter says what must be true; the framework makes it true at every decision and shows that it did.
Which is where the map meets the only question that matters to the person who signs. What you are buying, if you buy anything at all, is the ability to trust the AI your company already uses, shown with evidence, and a picture of your business that stays yours whichever vendors come and go. The map shows you, cell by cell, where that trust is earned and where it is still to be earned.
I keep one of those vendor diagrams in a drawer. It was a good diagram. It was also one slice, drawn by people who had no reason to draw the rest, and it took me a long time to notice that the problem was never the slices. The problem was that nobody had handed me the map they would all fit on.
If you want to go further, the choice is the one the front door asks, and it is the only choice here: I run a company, I advise companies, or I want to build my own. Choose yours.
The question I am still carrying
If your main AI vendor disappeared tomorrow, how many days until you were running again, and who would decide? It is the question that day of briefings should have made me ask, and nothing on my whiteboard could have answered it.
If you run a firm, Ground, MortarCloud.ai’s first engagement, starts with one decision: https://mortarcloud.ai/engagement/