The automation boundary in any workflow sits where automated work ends and manual work begins. That boundary has always existed, be it with spreadsheets or enterprise software, but it was skewed towards the manual side.
AI is pushing the boundary further and faster towards automation. It now reaches complex work that spans several systems and needs judgement.
The boundary does not land in the same place for every workflow. Call the place it lands the settling point. The business sets it by deciding how far to trust AI with each task.
Automation is a spectrum, not a switch
The standard view treats automation as a destination. Either you have automated a process or you have not. The conversation jumps from 'we need a human for this' to 'AI can do this now', as though the switch flips once and stays.
In practice, automation is a spectrum. AI can now handle use cases that previously needed human involvement at every step.
Switch
Spectrum
Rules-based systems handled repetitive, well-defined tasks. AI can take on work that involves ambiguity and interpretation. An AI agent can read a pricing system, a customer record and an order system in one go. It can reason across all three, recommend an action and then carry it out.
Together, these push more complex work past the point where it needed a human at every step.
Different workflows, different settling points
Copilot, human in the loop or human on the loop
The easiest way to describe each point on the spectrum is by what the human does there.
| Point on the spectrum | Who does the work | What the person does | Customer service example |
|---|---|---|---|
| Copilot | The person, with AI helping | Does the job and decides what goes out | AI suggests replies, finds answers and writes up the notes afterwards |
| Human in the loop | AI, up to a sign-off | Approves before anything goes out | AI proposes a refund, and a service agent signs it off |
| Human on the loop | AI, end to end | Watches, and steps in when needed | AI handles conversations alone while a supervisor monitors and can take over |
In the loop works like a second signature on a cheque: nothing goes out until a person signs. On the loop is a lifeguard: the swimmers get on with it, and the lifeguard only jumps in when something goes wrong.
Wherever AI does the work, it also needs a handoff: the moment it cannot finish a customer's request and passes it to a person. Done badly, the customer repeats everything. Done well, the person sees what was asked, what the AI tried and why it stopped.
Two platforms I built settled at different points on this spectrum.
The first was a customer onboarding platform for a global B2B business. Onboarding meant collecting documents, verifying identities and running regulatory checks across 100+ countries. It took four days. With self-service document uploads and automated AI-powered verification, it came down to under four hours.
The workflow moved dramatically towards full automation. But a human still reviewed the exceptions: documents that did not match, flagged customers, edge cases where the regulation was ambiguous.
The system cleared the routine cases on its own. Its settling point sat close to the automated end, with a handoff for the exceptions.
The second was a pricing platform for the same business. AI combined product, customer, route, capacity and order data to recommend prices as conditions changed. Commercial teams accepted or overrode every recommendation.
Its settling point sat in the middle of the spectrum, with the human in the loop: AI recommends, the human decides, the system acts. That was by design. The cost of a wrong price on a strategic account outweighed the efficiency of full automation.
This is a product and business decision. Not a technology one.
Customer service shows why one settling point cannot cover a whole workflow. A password reset and a large refund sit in the same workflow. They should not sit at the same point on the spectrum.
The reset can run end to end with a lifeguard watching. The refund needs a second signature.
The settling point differs by workflow, and by task.
Customer onboarding checks
AI handles: Document checks, identity verification, regulatory screening
Human handles: Flagged customers, ambiguous regulations
Pricing
AI handles: Combines the data, recommends prices
Human handles: Accepts or overrides every recommendation
Customer service: password reset
AI handles: Verifies the customer, resets access
Human handles: Watches, steps in if something looks wrong
Customer service: large refund
AI handles: Gathers the facts, proposes the refund
Human handles: Approves or declines before money moves
Same workflow as the password reset, different settling point
Illustrative. Settling points vary by organisation and industry.
Design the roles around the decisions AI hands back
As AI takes on more of the routine, the human role does not shrink. It changes.
The people who used to handle the full workflow now concentrate on the cases AI hands back: the override, the escalation, the judgement call.
On the pricing platform, commercial teams stopped building routine quotes by hand. They reviewed the recommendations and spent their time on strategic accounts where the context was too nuanced for an algorithm. On the onboarding platform, compliance analysts stopped chasing documents and focused on flagged cases that required investigation.
Personal AI agents are starting to push the same shift from the customer's side. When a customer's own agent can rebook a flight or arrange a return, that contact never reaches a person. What does arrive is the case the machine could not settle: the dispute, the bereavement, the fraud, the high-value account.
The work left is harder and more human, and it needs judgement and the authority to act. The role does not change by default. Someone has to redesign it on purpose, or it drifts.
The settling point will keep moving. Better AI is only part of the reason.
Trust moves the settling point, and it is earned one task at a time. A task that needs a second signature today can be left to a lifeguard once the AI has a track record. The human role shifts with it, from handling one case at a time to overseeing many.
But the human will not disappear from the spectrum. The human moves to where the stakes are highest and the context is most nuanced.
The settling point is different for every workflow. Set it deliberately by asking these three questions of each task:
- What does a wrong answer cost?
- How much context does a good decision need?
- What track record has the AI built on this task?
A high cost or deep context keeps a person in the loop. A low cost and a long track record let the person step back and keep watch.