Most people use AI. Few have changed how they work.
Our research follows one gap: where it comes from, what closes it, and what changes in the work when it closes. We publish the scoring method with the findings.
Method and score calculatorMany footprints stop at the water. Few cross the stones.
Our research has three studies, eight notes and five field notes.
Each title below is the finding of its piece.
Eight notes describe how agent systems are built.
- Six professionals who do not code organized their agent workspaces the same way.
- Every AI tool has a point where it stops working, so choose by stakes.
- AI governance is a data infrastructure problem.
- Capture every correction in a skill, so you never teach the same lesson twice.
- A team brain loads in three layers: the institution, the person, then live context.
- A channel is a door, not the agent.
- One agent is the start: sub-agents and a dispatcher let many run at once.
- A blueprint for multi-agent systems that improve themselves.
Five field notes record what we saw in the work.
- Consultants use ChatGPT for research and brainstorming, then stop.
- Firms handed out prompt libraries while the tools that do the work went unused.
- Progress with AI comes from context, skills and memory, not from a better model.
- Structured context beats better prompts.
- People learn agents by building one skill from their own work on day one.
Most regular AI users score in the lower half.
We scored 54 interview transcripts on AI capability. The median was 38 out of 100. Median scores fell from 55 in everyday chat to 12 in automation and agents.
Read the studyOn a tidal rock, mussels and barnacles crowd the lower half and thin out toward the top.
Confidence in AI runs far ahead of capability.
Confidence averaged 66 out of 100 and capability 37. Confidence ran higher in 53 of the 54 transcripts.
Read the studyAt low sun, a small tree casts a shadow many times its own height.
More than 100 interviews show where people get stuck.
Who we spoke to, how we score what they say, and the 212 pain points coded from the first 38 interviews.
Read the studyPressed plants, one to a sheet, sorted by kind.
Working on the same question?
We look for people to interview, data to compare against, and teams willing to test what we find. Leave a note and we will write back.