One Ghost, many canvases

For the past year I have been building the pieces separately. A cockpit to watch the fleet. A discovery loop that hunts every morning. A measurement loop that checks how we show up when AI answers questions about the categories we work in. Each one worked. Each one was its own app, its own screen, its own habit.

A few weeks ago the pieces resolved into one idea, and I have not been able to see them any other way since.

Ghost is the intelligence. The canvas is the view.

Ghost is one platform: a single intelligence fabric underneath, and canvases on top. A canvas is not another app with its own database. It is a view into the same fabric. A person, a piece of evidence, a workflow that appears on one canvas is the same object everywhere it appears, carrying the same history and the same confidence.

The canvases we run today: Discovery, which watches what is moving in a market and finds the warm path to the conversation. Visibility, which measures whether you show up when AI answers the questions your buyers actually ask. And Workflows, the work in motion, the gates that need a human, and what changed because of what shipped.

Ask Ghost

The doorway is a question. You do not pick a model, route an agent, or choose a tool. You ask Ghost, and the answer materializes on the canvas as objects you can enter: an investigation, a comparison, a plan, a saved view. You can trace why Ghost believes what it told you, down to the source and the date. Ghost proposes the next action. You approve the ones with consequences.

Every question leaves something behind. Every action leaves a trace. Every outcome changes what Ghost knows.

Not chat painted beige

The failure mode I am steering hard away from is a chat window with a nicer palette. Chat is transient. You get a good answer, the scroll moves on, and the answer is gone. On a canvas, answers become objects, the objects stay put, and the map is still there tomorrow. You develop spatial memory for your own company. The world does not reshuffle itself every time you ask.

And the agents stay backstage. Nobody should have to know which model ran or which agent handed work to another agent. You should experience the outcome: I asked Ghost. Ghost showed me what is happening, why it matters, and what I can do next.

The canvases were not the point either

I wrote this in July and called it a rethink. It was one, but I had the altitude wrong. Collapsing three tools into one surface is an interface decision, and interface decisions are the sort of thing a competent competitor copies inside a quarter. If that were the whole idea it would not be worth the year it took.

The part that matters sits a layer lower. The software stack is finished except at the top. Systems of record store what happened. Systems of work move it through boards and pipelines. Systems of action showed up with agents and do the work. Above all of that there is supposed to be something that turns outcomes and judgment into capability that compounds, and almost nobody ships it.

Models reason. Agents act. Nothing owns the decision.

That is what the fabric underneath the canvases actually is, and it took me until August to say it plainly. Not storage, and not a nicer window onto the agents. A record of what got believed, what evidence earned it, who agreed or overruled it, and what happened next. The canvases are how you look at that record. The record is the thing.

Which also settles what to do when the model changes underneath, because it will. A canvas built on a vendor dies with the vendor. A record of your own decisions does not care which model did the reasoning this year.

Where this actually sits

The cockpit that started this runs my portfolio today. The discovery and measurement loops run daily. The unified canvas is specced and in build. What I have not earned yet is the second half: outcomes at volume, and a lift I can attribute to them rather than to a better model arriving underneath me. I am writing the destination down because naming it is how I hold myself to it.

The argument for why the record beats the model is Evidence compounds, and what happens once it gets good enough to act on is When the loop closes. The machine underneath is still The Loop Is the Product, and the two loops feeding the first canvases have their own pieces: the hunt that runs before I wake up and why one AI answer is not a ranking.

One fabric. Many canvases. One record underneath all of it.