Contents · 7 min read
01 / 05 · Time saved
When is an agent still a productivity tool?
In short
An internal drafting agent may help your team finish work sooner. That gain matters because staff can use the saved time elsewhere.
You see an agent-written draft ready for review, and your team finishes the customer file sooner. But the customer still buys the same result through the same journey, and the work after the draft still needs the same staff effort, so the agent has saved time without changing the business model.
Your internal work may take less time without changing the offer. The agent helps your team, but it has not yet taken on part of what the customer buys.
As Copilot Is Not an Agent explains, an agent applies your business context and a set method to a task. Several agents can also work in order when the task has more than one part.
That can save many hours, which is a sound result. To find a business-model change, you must now look at what reaches your customer.
02 / 05 · Customer value
When does the business model start to change?
In short
The change starts when an agent performs part of what your customer buys or changes how your company delivers it. It also starts when serving one more customer needs far less added work.
Begin with the offer your customer sees. Write down what the customer buys today, then state what the agent does within that result.
An agent that only helps an employee prepare has not crossed the line. An agent crosses the line when the work it completes is part of the result the customer pays for.
The same test applies when the customer does not see the agent. What matters is whether the agent changes the work needed to produce the result.
This is why faster drafting is only the floor. The AI-augmented engineering organisation shows that agents can inspect and test bounded changes, while people still own the judgment that puts them into use.
The shift can go further when AI takes on work that once drove the fee. The professional services CEO briefing shows why that can force a change in how the firm earns money.
Once an agent performs customer work, the next issue is who lets it act and who answers for the result.
03 / 05 · Control
What control does execution require?
In short
Execution needs tighter control because an agent can now change records or prepare actions that affect real work. Your company needs clear access rules, approval points and one named person who remains accountable.
support case within set limits Action Each step gives more access
The move from advice to action happens in steps. Each step gives the agent more access, so each one needs a clear human decision.
Your team chooses the material
The agent works from material your team has chosen. It can advise, but it cannot read current system records or change them.
Read access
The agent reads approved records from your systems. The named owner still decides what follows from what it finds.
A person approves the change
The agent drafts a bounded change for a person to approve. The person checks the result before it reaches the live system.
The agent may create a support case within set limits
The agent completes preset actions within agreed limits. A named owner remains accountable for those limits and for any result with real consequences.
The final check is control. State which records the agent may read and which actions it may prepare. Then set which actions need approval and what stops the agent in an emergency.
The AI-augmented IT department sets out why IT must control identity and access. It must also keep records of agent actions and the limits around them.
Each month, the CIO or CTO should review denied actions and human overrides. The same review should cover any incident to which an agent contributed, because those cases show whether the limits still work.
People can keep the choices with real consequences while agents prepare bounded work. Once that pattern is safe, you can test whether it works across more customer work.
04 / 05 · Service capacity
When does agent work change service capacity?
In short
Service capacity changes when controlled agent work can repeat across more customer work without the same rise in employee work. One successful case is useful, but the business model changes only when the pattern holds as volume grows.
A one-off agent may save time on one piece of work. That is a good place to learn because your team can see where review is still needed.
If each new request needs a new system link or a fresh set of rules, each added customer still creates setup work for your team. Repeated work can become an approved way for agents to read or act, which lets the same control support more customer work.
The AI-augmented IT department paper describes how repeated links can become approved tools. Your teams can then use them while IT keeps control of access and records.
Now compare the added customer volume with the added employee hours needed to deliver it. If each new customer still needs the same human work, your capacity has not changed much.
If the agent repeats a bounded part of delivery, more work may move through the company without the same rise in staff effort. That change can alter the cost of serving one more customer.
You still need people where judgment has real consequences. The next choice is whether the proven change is strong enough to alter what you sell.
05 / 05 · The offer
When should you redesign the offer?
In short
Redesign the offer only when you can show a change in customer value or in the cost of serving more customers. If neither has changed under safe controls, keep the agent inside the company as a productivity tool.
An internal agent can still be worth using. It can save time and help your staff prepare better work, even when the offer stays as it is.
But a new offer needs more proof because it changes what customers expect. Your team should show what the customer gets and how delivery work changes.
Use this review with your leadership team:
Have sales state what the customer buys
Write one plain line for the offer today and one for the proposed offer.
Have the work owner show the agent’s part
Mark where it advises, prepares an action or completes bounded work.
Have finance test the added work
Compare added customer volume with the employee work needed to serve it.
Have IT state the limits
Record what the agent may read and what it may change. Name the person who approves each step.
Have management choose the offer
Change it only if the customer or the cost of added delivery has changed under those limits.
Do not redesign the offer around an answer from a general model because competitors can ask it the same question. I Run a Strategy Firm explains why the same model tends to give firms the same common answer.
Base the offer on the process the agent follows and the company material it can use. That material can include your strategy and numbers. It can also include what you know about customers and the commitments you have made.
The next decision
Your leadership team now needs to decide whether agent work has changed what customers buy or the cost of serving them, then set the access and approval limits that make the change safe. The Access Ladder shows how much system access an agent needs at each stage, from material your team selects to bounded action in live systems. Use it to match each proposed part of customer delivery with the access it requires before you redesign the offer.
How much access to your systems an AI agent needs, rung by rung. Free, with a PDF.
Questions CEOs ask about this
When do AI agents become a business-model change rather than a productivity tool?
AI agents change the business model when they take on part of what customers buy or let your company serve more customers without the same rise in employee work. If the offer and delivery work stay the same, the agent is a productivity tool, even when it saves many hours. Your decision is whether the proven change is controlled enough to alter what you sell.
What changes when agents can execute rather than only advise?
The agent can prepare or complete a bounded action instead of stopping at an answer. That gives it more effect on real work, so access rules and human approval become part of the design.
Which customer journeys can become agent-delivered services?
Start with a journey where the work is clear and the agent can stay within set limits. No single list fits every company, so test the customer result and the needed human judgment in your own.
How do agents change the cost of serving more customers?
They change it when a repeated part of delivery no longer needs the same rise in employee work. If every new customer still needs the same human effort, the cost of added delivery has not changed much.
What access and controls are required at each stage?
Advice can begin with material your team selects. Reading live records needs approved access, while prepared actions need a named approver. If the agent may change live records, it also needs its own identity and action records, plus a way to stop it.
Which agent advantages are defensible?
Access to a general model is not enough because a competitor can use the same model. A stronger advantage comes from your company context and method, held inside work that you own and can control.
Founder, GRAIL
Johan has spent fifteen years in management consulting and executive leadership, including as Chief Product Officer of a SaaS company. He founded GRAIL to help European leadership teams work out what AI changes at the core of their business. He is the author of The Execution Revolution and the host of the ThinkRoom podcast.
Related reading
Sources
- Copilot Is Not an Agent: what an agent adds beyond a general model. (accessed October 29, 2026)
- What does an AI-augmented IT department look like?: how agents read and act through approved access. (accessed October 29, 2026)
- What does an AI-augmented engineering organisation look like?: how agents prepare bounded work while people keep accountability. (accessed October 29, 2026)
- Professional services CEO briefing: why AI can change the economics when it takes on paid production work. (accessed October 29, 2026)
- GRAIL vs the alternatives: why agents can remain on the client’s accounts and belong to the client. (accessed October 29, 2026)
- I Run a Strategy Firm: why access to the same general model does not create a lasting edge. (accessed October 29, 2026)
Next scheduled review: February 5, 2027