Issue 1 · July 14 · Issue 2 · July 21 · Issue 4 · August 5 · Issue 5 · August 11 · Issue 6 · August 19 · Issue 7 · August 26 · Issue 8 · September 1 · Issue 9 · September 8 · Current issue
A weekly read on what the AI noise is hiding.
Last week the CEOs admitted AI had not paid off yet. This week OpenAI published what people actually do with it, across 800,000 work messages. The headline finding is about reaching outside your own job. The finding underneath it is about what stops you, and that one is about the size of the company you work for.
What the AI noise is hiding. Every stat sourced. The hype named as hype. One number worth acting on. Read the full brief below, free, no email required.
The report’s subtitle is the part most people will skim past. How AI is expanding what people do at work. Expanding. Not automating. Those are two different projects, and most companies have not decided which one they are running.
OpenAI published a study last week that measured something worth arguing about.
They looked at 800,000 work messages and asked a simple question. When someone asks AI for help, is the task part of their job, or does it belong to somebody else’s?
Of the messages tied to a specific occupation, 43.5 percent were work that historically belonged to another job. A salesperson running the analysis. A marketer troubleshooting the site. A designer writing the contract summary.
That number is worth a pause. Close to half of the occupation-specific work people bring to AI is work their job description does not cover.
Then there is the finding that is easy to miss. The bigger the organisation, the less of it happens. Among typical users, boundary crossing runs at 18.9 percent in workspaces of two to five people. It falls at every size step. By 101 seats and up it is 16.3 percent.
OpenAI’s own reading of that is the sentence worth going back to. In a small company a person drafts the copy or reviews the contract themselves, because there is nobody to hand it to. In a larger company, they rely on an established team, workflow, or internal service.
Read that again as an operator. The capability did not change. The same person, the same tool, the same task. What changed is that the large organisation has a process for it, and the process is where the reaching stops.
The consensus has been circling this for a year without landing on it.
Gartner’s entrenchment research puts a ratio on the same problem from the other direction: 100 days of AI implementation can require up to 200 days of change management. The technical work is the short half. Gartner goes further than most and admits that standard change management can make resistance worse rather than better.
BCG’s split from Issue 2 still holds and is worth repeating once as a callback rather than as news: 70 percent people, 20 percent process, 10 percent algorithm.
Put those beside the OpenAI gradient and they describe one thing from three angles. Nobody is short of models. What differs between organisations is the machinery wrapped around the person using one.
Here is where honesty costs something.
The size gradient is a 2.6 point gap. It is not a canyon. The study is descriptive and cannot prove cause. OpenAI is measuring its own product’s usage, on a sample the report itself says does not represent the whole workforce. Workspace seats are not company headcount. Anyone selling a transformation off this chart is selling something.
What it does support is narrower and still useful. People in larger organisations reach outside their own job less often than people in small ones.
That is the whole finding. It is smaller than the story most people will tell about it this week, and it is still the most interesting thing in the report.
Which brings it back to the subtitle. Expanding, not automating.
Automate the task and you take work off a person. Expand the person and they walk into rooms they could not walk into before. Same technology, opposite outcome. The choice gets made deliberately, or it gets made by drift.
If your organisation has a team, a workflow, or an internal service for everything, you have already chosen. You just did not do it out loud.
And the same gap as every week still sits underneath it. The datasets are enterprise-scale. The prescribed fixes assume a budget most operators do not have. The $10M to $250M company, the one that feels coordination drag hardest and carries the least legacy weight to fight, is standing on unclaimed ground.
2.6 points.
Not 43.5. The big number is the one that will travel this week, and it is real, but it mostly tells you that people use AI to reach past their job description, which anyone who has watched a marketer debug a website already knew.
2.6 points is the gap between how much reaching happens in a five-person shop and how much happens in a company past a hundred seats. It is small. It is stated here as small, on purpose, because the temptation is to inflate it into a law about big companies and it does not carry that weight.
But it points at something real. The process you built to make the work reliable is also the thing standing between a capable person and the work next door. That is not an argument against process. It is an argument for knowing which of your processes is still earning its cost.
You do not need a study to find out. Measure the coordination tax you are paying right now. That is what the Agentic Readiness Score does. 33 questions, free, and it ends with a real number.
One weekly read on what the AI noise is hiding. Every stat sourced, the hype named as hype, one number worth acting on. No spam, unsubscribe anytime.
Start by measuring the coordination tax you’re paying right now. Free, a few minutes.
Next Tuesday: another scan, same rule. Cut the hype. Show you what the noise is hiding.