There is a moment that happens at almost every company I sit with. It usually comes about ninety minutes in.
The person across from me is the one who keeps the place running — the one everyone goes to when something is on fire, who knows which projects are slipping and which clients are quietly unhappy. We have been walking through how a proposal actually gets put together at her company. Who pulls past project data. Who writes the price. Who reviews the technical sections. Who clicks send.
Somewhere in that walk-through, she stops. And she says some version of what I now hear from almost every company I work with:
"I've never written any of this down. We've been running this way for years."
A few weeks ago, Jack Dorsey and a partner at Sequoia published an essay arguing that AI is now good enough to do something companies used to need a lot of people for: keep track of what is happening across the company. Block laid off 4,000 people on that argument. They were not the first to make it.
Most of the companies I work with cannot run that play yet. Not for the reason you'd think.
What the whole industry is saying
The Block essay is the most prominent version of this argument, but it is not the only one. Read three pieces from the last year side by side and you start to see the same conversation happening from three different angles.
Block, from the inside. Companies got too big for one person to see, so we invented layers of managers — and the work of those layers, more than anything else, was passing information. Who's working on what. What's stuck. Who knows the answer to this question. Block's argument is that AI can now do most of that work. Keep a current picture of what's actually happening, surface what matters, route the right question to the right person without needing a chain of managers in between. They are not pretending those jobs disappeared. They built a system that does that work instead.
Theory Ventures, from the capital side. A few months earlier, a venture firm called Theory published an essay arguing that the missing piece for enterprise AI is what they called a "living system of record" — a continuously updated picture of how a company actually operates, including all the unwritten rules. Their bet: the next major category of business software, on the order of accounting or CRM, is the layer that watches the work, learns the rules, and feeds context to the AI when the AI tries to help. They expect billions of dollars to go into building it.
MIT, from the data. Last year, researchers studied 300 enterprise AI deployments. Ninety-five percent of them produced no measurable result. The reason, almost every time, was not that the model was too dumb. It was that the AI didn't have enough context about the business to be useful — the rules were missing, out of date, or contradictory. Gartner formalized the shift a few months ago: figuring out what the AI needs to know to do its job is now the work. The model is not.
The Theory essay puts the diagnosis plainly: "Operational documents are incomplete, outdated, and even contradictory."
All three of these conversations are happening at the top of the market. Block can do this. They have the engineers, the data from millions of Cash App and Square transactions, and a decade of building software that watches itself. The companies the AI vendors are selling context platforms to look more or less like Block.
Now the harder question. What about everyone else?
Why most companies cannot run this play
Three reasons. None of them are about how good the AI is.
There is nothing for the AI to observe. A "living record" of how a company operates needs activity it can read. At Block there are logs, transactions, tickets, screen recordings, ten years of structured work product. At most 50-to-200-person companies, the operation is verbal. How proposals get put together, how a project moves from sale to delivery, who reviews quality, how a change gets routed — most of it lives in habit, not in any system. There is no log to ingest. No ticket trail. No screen recording. The work happens, then disappears. Even the smartest context platform on the market would have nothing to learn from.
The "managers" in those essays are not the same as the people in the middle of a 90-person company. At Block, a middle manager was mostly a router — someone who passed information up and down a chain. At a company of 60 to 200 people, the people in the middle do something more complicated. They hold relationships with clients. They decide which work to chase. They have taste — a sense for which approach is right, which scope is fair, which hire is going to work out. AI can route information today. It cannot, today, hold the taste of someone who has run a hundred big proposals.
The bottleneck is not the org chart. It is not the AI either. Most companies don't have a hierarchy problem or an AI problem. They have a visibility problem dressed up as both. Adding a layer won't fix it. Removing a layer won't fix it. Buying a context platform won't fix it. None of those work, because the underlying work has never been drawn. You cannot fix what you cannot see.
AI without operational clarity is just expensive noise.
What "drawing the operation" actually means
The phrase sounds abstract. It is not. It means six concrete things any leader recognizes once you name them.
- Who actually does what. Not the org chart. The real version. The person who quietly owns quality. The one everyone asks when the spec is unclear. The one who edits every proposal before it goes out. Often these are not the people their card says they are.
- How the work actually moves. A project from the day someone calls in to the day the invoice gets paid. Every step. Every place it sits. Every place it changes hands.
- How decisions get made. Who decides what. How long decisions take. What it costs every time one slips. For most companies, this is the most expensive thing they don't track.
- Where the knowledge lives. Whose head. Which folder. Which old project file. What disappears the day that person leaves.
- Where the work changes hands. Sales to delivery. Design to build. Senior to junior. These are the places margin leaks. Not because anyone is bad at their job, but because what one side knows, the other side has to be told — and usually isn't.
- What the company sees about its own work. What gets tracked, what doesn't. Where the numbers come from. Where they are made up. At most companies, the weekly report is part real, part guess. The company runs on it anyway.
A company that has done this work has the start of what the AI essays are calling the missing layer — a living, current picture of how the work actually moves. A company that hasn't done it doesn't, no matter how good the AI gets. The order matters. You can't hand AI a company it can't see.
Across the companies I have sat with, the most common finding is not "your team is overworked" or "your tools are wrong." It is that nobody at the company has ever drawn how their own work moves. The person who keeps things running has a piece of it. The owner has a piece of it. The project leads have pieces. Nobody has the whole picture.
Why Sterity exists
Sterity is named for what gets left behind. Posterity. The system that still works in 2030 — after the team has changed, after the AI has changed, after the market has changed.
We start with a free Health Check. About ten minutes, online, no sales call. It is the smallest possible first sketch of the operation — what one person sees of the company, named clearly. Most people find it more useful than they expected.
When companies want to see the fuller picture, the next step is to bring in the rest of the people who hold the other parts of it — the person who edits every proposal, the one who reviews quality, the one who keeps clients close. The picture only emerges when more than one person draws it. We give your team a private space to do that work together, and we sit alongside it.
When companies decide they are ready to fix what the picture shows them, we map the operation across the six pieces above and build the few tools that fix the things quietly costing the most. Every tool we build, the company owns. We measure success by how quickly they stop needing us.
Proprietary tools, not proprietary dependency.
The tools we use to find what's broken are ours. The tools we build to fix it are yours.
The honest tension
If Block is right, and the investor essay is right, won't this work get easier? Won't AI eventually walk into a company, watch how things actually run, and draw the picture itself?
Probably, yes. The AI industry agrees with this. Theory Ventures' bet is that within a few years, software will exist that learns how a company operates by watching the work happen — and that this software will be the foundation every other AI tool runs on top of. The Health Check that takes ten minutes today will take thirty seconds in 2028. We are building toward that.
But the companies that benefit from that future will be the ones that started now. AI does not arrive at a company and pull the operation out of thin air. It needs the operation to exist in a form it can read. Verbal, in the air, lived only by the people who built it is not a form anything can read. Companies that wait will be five years behind companies that started.
The technology is not the bottleneck. The discipline of looking is.
The order matters
The person I mentioned at the start of this — the one who had never written down how a proposal actually got built at her company — finished that conversation and said: "Now that I can see it, I can't un-see it." That is the moment we work for.
Most companies don't need an AI strategy yet. They need to see their own operation, on the things that are quietly costing them money. When that is in place, the AI strategy is obvious — it falls out of what the picture shows you. The order matters.
If any of this resonates, the Health Check takes about ten minutes and is free. No sales call required. If something it surfaces is worth a longer conversation, we'll have one. If you want to bring more of your team in to see the fuller picture, we can do that too. If not, you keep the report.
Sources referenced in this article
- Jack Dorsey and Roelof Botha, From Hierarchy to Intelligence, Block, March 2026.
- Theory Ventures, The Business Context Layer, September 2025.
- MIT NANDA, State of AI in Business 2025 — the 95 percent figure on enterprise AI pilot outcomes.
- Gartner, July 2025 declaration on the shift from prompt engineering to context engineering.