AI agents that run real business processes now work, and they are going to redraw markets. Which side of that a business lands on will be decided by leadership, not technology. Specifically: understanding how the organisation really works, and having the courage to redesign how work flows before you automate it. Without that redesign, AI adds value in spots, and sometimes subtracts it. With it, a business can operate at a level it could not reach before.
The agent patterns have settled
For most of 2024 and 2025, building agents meant exploring. Frameworks came and went, demos impressed and broke, and every team was working out the shape of the thing for itself.
That period is over. The winning patterns have been discovered, tested and proven. An agent that runs a process now has a recognisable design. It watches the inbox or portal, reads what arrives, matches it to your records, decides within rules someone has written down, acts, and escalates to a named person when it should. It leaves an audit trail on every case. You can see that shape in claims intake and in credit control, and in late September the large platforms put always-on agents in reach of small firms for the price of a software seat.
So building the agent is no longer where projects stall. The question has moved to which work the agent should run, and what has to change around it first.
The hard part is knowing how your business really works
The org chart tells you who reports to whom. It does not tell you how work actually flows. Which approvals sit for three days. Which spreadsheet is the real system of record. Which team quietly rekeys another team's output. Which person everyone goes to because only she knows how the exceptions are handled.
Those are the dynamics, the choke points, the conflicts and the key nodes in the flow of work, information and decisions. Redesigning them to remove friction and let work flow is change management and process redesign, and it has always been leadership work.
When we built the agent that went on to resolve 67% of customer service cases at one of Europe's largest insurance brokerages, we started with the process and the data, not the model. We mapped every decision point, chose which ones could be delegated to the system with acceptable error rates, and designed the escalation so the rest reached a person with full context. The team was talented. The problem was never effort.
What redesign actually involves
In my experience redesign comes in five moves, and the order matters:
- Reorganise so your best people contribute more. Move them to where their judgement changes outcomes, and take the routine work off them.
- Gather more and better data, so decisions are better informed rather than argued from memory.
- Strengthen meritocracy. Make it visible whose ideas, decisions and actions actually lead to the best results.
- Learn what works and spread it. A pattern that works in one team should become the way the whole business works.
- Then scale, amplify and automate.
Automation is the last move, not the first. An agent pointed at a broken process runs the broken process faster and around the clock.
McKinsey found the same. Its State of AI 2026, published in August, reports that nearly three-quarters of the highest performers, about one company in twenty, have fundamentally redesigned their workflows around AI, up from 55% a year earlier. Among everyone else it is one in four. The same high performers are twice as likely to say their senior leaders visibly back the work.
AI supports the redesign. Leaders do it
AI can help a great deal with the analysis. It can read a year of email and show where work waits, map how long each step takes, and spot the cases that keep bouncing between teams. I use it for exactly this.
But the redesign itself needs context that is not written down. Who really makes a decision. Why two departments do not trust each other's numbers. Which change a team will absorb this quarter and which one will break it. That is tacit knowledge, experience and intuition about people, and it is not in any system a model can read.
So the decisions stay with leaders. Moving a team, changing who approves what, or retiring a step that someone built their role around, and moving them to work that needs their judgement, takes vision, clarity of intent, courage and a willingness to own the result. Those decisions cannot be handed to a model, and they should not be handed to a committee either.
AI amplifies whoever holds it
Without redesign, AI still helps. It helps in spots, and it helps unevenly. McKinsey's 2026 survey puts numbers on it: 80% of respondents say AI has made them personally more productive, but only 37% report any effect on operating profit, barely changed from the year before.
Give it to a driven, curious person and they use it to improve the speed and the quality of their work, often dramatically. Give the same tool to someone who would rather do less and they will use it to do less, and hide behind output that looks finished. Same tool, opposite results. That is how AI ends up adding value in one team and subtracting it in the next. The difference is the operator and the system around them, never the technology.
The same is true of whole organisations. Clear rules and people who own outcomes get multiplied. So do workarounds. Redesign is how a leader decides which of those gets amplified.
The market is about to be redrawn
When a business redesigns itself around agents, grounded in a real understanding of how it works and backed by sound decisions, the result is not a few percent of efficiency. It can operate at a different level, growing without hiring ahead and putting its best people where their judgement counts.
That changes the rules and the balance in a market. Smarter new entrants that build this way from day one will pull ahead, and so will the established firms that reinvent themselves best. Some of them will come to dominate. The ones that do not will find their weaknesses more visible and their ground harder to hold, and they will lose it faster than they are used to.
None of this is decided by who buys the best model. Everyone can buy the same model. It is decided by who does the redesign well.
The qualities that decide the winners
In the age of AI, it is the most prized human qualities that will separate winners from losers. Judgement. Clarity. Vision. Determination. The ability to inspire others. Intellect. Unbiased, sound decision-making. That is great leadership.
I would add two more. It has to be selfless, because good redesign often moves power away from the people who currently hold it, including leaders. And it has to be merit-based, because the point is to see clearly what works and who makes it work, and to back them.
I learned the personal version of this years ago, when I realised that the engineering was the easy part of my own job. Agents are about to teach it to whole organisations.
If you want to work out which process in your business is ready for an agent, and what has to change around it first, that is what a Process Audit is for. Let's talk.
Related: After 17 years as a CTO, I learned the engineering was the easy part · Always-on AI agents are here. The agent is now the easy part · Stop counting AI use cases. Redesign three value streams instead · Leaders are measured with a different stick
