Forever Foundry Book a scoping call

Blog: AI agents

Your AI agents may all be waiting for the same person

Published by Scott Seaborn, Founder

Finance automates invoice checks. Procurement automates supplier reviews. Operations automates order exceptions.

Each proposal includes a reassuring line: a person will review anything uncertain.

Then all three send their exceptions to the same busy manager.

Human involvement remains part of the work

On 9 November 2026, Zapier published new survey findings. Among leaders whose organisations use AI agents, only 30% said the technology handled most of its tasks without human intervention.

Centiment surveyed 513 US managers and executives at organisations with at least 1,000 employees between 16 and 21 September. These are reported experiences from large US organisations, not a benchmark for every business. Read the research.

The FOFO™ view

Human review is a service that needs capacity, authority and a response time.

Imagine a property services company introducing AI to triage maintenance requests. Routine jobs move quickly. Unusual repairs need approval from a regional manager who already handles complaints, contractor queries and urgent site issues.

The system runs overnight. The manager does not.

An individual pilot might look successful because its small queue receives special attention. Rolling it out across every property could create a backlog that customers experience as silence.

Before expanding, ask reviewers what else competes for their time. Ask frontline teams how long decisions can wait. Check whether several proposed automations depend on the same people.

That shared demand belongs in the investment decision.

Budget the review capacity before expanding

Estimate the number of cases needing intervention and measure the time to resolve them. Test peak periods as well as averages.

For each review queue, agree:

  • Who has the knowledge and authority to decide?
  • When are they available, and who provides cover?
  • Which cases take priority?
  • What happens when the queue exceeds capacity?

Track waiting time, overdue cases, review effort and decisions later corrected. A low intervention rate can still create substantial work at high volume.

The next step might be a smaller rollout, better information at intake or additional review capacity. Removing necessary checks simply to clear the queue would undermine the design.

AI Compass™ gathers insight through AI-powered surveys, branded for your organisation and sent by you. Slid3rs™ helps teams prioritise opportunities by business value and delivery complexity, including the human work required to make them succeed.

Before funding another agent, establish who will be there when it needs an answer.

Related reading

Next

Understand the workload behind your next investment.

Explore AI Compass™