In 2025, Microsoft named a new kind of organisation: the Frontier Firm. It described companies built around human and AI agent teams, with every employee becoming an "agent boss".
A year later, the 2026 Work Trend Index shows how far that has gone. Microsoft surveyed 20,000 AI users across ten countries and analysed Microsoft 365 usage signals.
How many firms are really at the frontier
Only 19% of AI users fit Microsoft's Frontier profile. Another 10% say they are actively blocked from using AI well.
Agent use is rising fast. Active agents in Microsoft 365 grew fifteen times in a year. But most people still work in organisations that were designed before agents arrived.
What Frontier teams do differently
The striking finding is how ordinary the differences are. Frontier teams were far more likely to share AI mistakes openly, 61% against 36%.
They were also more likely to discuss quality standards for AI work, 54% against 29%. These are habits of conversation, more than technology.
Microsoft also found organisational factors explained 67% of reported AI impact. Individual factors explained 32%. The team around a person matters more than the person's prompting skill.
Build this capability in your organisation
Growth Performance designs bespoke programmes that turn insight into measurable behaviour change.
Agents still need people to join up the work
Directing agents is a real skill. It sits on top of the human skills that already matter: clear goals, sound judgement about quality and noticing when something is wrong.
Agents are weak at coordinating with each other. The 2026 CooperBench study found coding agents were 30% less successful when working together than alone. Messages were vague and commitments slipped.
Human and AI teams also change how people work. In a 2026 experiment, Ju and Aral found human-AI teams produced more output but more similar output. People sent fewer interpersonal messages.
What this means for leaders
The Frontier Firm is an operating model. It asks leaders to work on three conditions at once.
People must take on the change and make agents part of their work. That is Adoption. People must be able to direct, combine and check agents. That is Adaptation.
The organisation must stay ready as agents keep improving. That is Acceleration.
Try this
At your next team meeting, ask everyone to share one AI or agent mistake from the past fortnight. Go first yourself.
Then agree one quality standard for AI-assisted work. Write it down and review it in a month. The Checking Capacity Calculator helps you plan the review time.