When the draft is fast but the work is still waiting
A sales team can generate follow-up drafts quickly and still finish the day with an untouched review queue. Someone has to check what was promised, resolve missing context, approve the message and make sure the next action happens.
That is not an argument against AI. It is a reason to look beyond the step that became faster. The work may have disappeared, moved to another person or returned in a different form.
What Anthropic's October research actually reports
Source findings: In its 16 October 2026 study, Anthropic reports that employee-directed interactive agent runtime reached about six times estimated employee work hours by August. The comparison includes subagents; employee hours assume a 40-hour week. Runtime excludes waits longer than 30 minutes.
A separate measure found roughly 80% of interactive session wall-clock time was spent waiting for human input. These are different denominators, not interchangeable measures of productivity. The study also reports output gains alongside pressure on human attention.
Limits: This is one unusually AI-native company studying itself. It does not establish a net productivity loss, identify all waiting as wasted time, or provide ratios that ordinary SMEs or European companies can safely adopt. Read the research and its definitions.
Where the extra human work appears
The Leading Space interpretation: The useful management question is not whether the agent stayed busy. It is whether people can complete more worthwhile work at an acceptable standard.
- Review: someone must establish whether an output is usable.
- Rework: a plausible draft may need missing context or substantive correction.
- Coordination: the next owner needs to understand what is ready and what remains unresolved.
- Exceptions: unusual cases need decisions, not another routine answer.
- Management: maintaining instructions, checking performance and resolving recurring issues takes time too.
Some of this is valuable work. Commercial judgement should not disappear simply because drafting became faster. The question is whether avoidable preparation and checking are consuming the capacity AI was supposed to release.
Do not fix the queue by removing judgement
If every draft needs the same attention, a senior person can become the bottleneck for work that did not previously reach them. But removing review indiscriminately can turn a visible queue into an invisible quality problem.
The better question is which decisions need a person, and what that person needs to decide well. A short brief with the relevant evidence and unresolved point can be more useful than another long summary.
The AI Review & Approval System addresses the path from prepared work to an accountable decision. The AI Attention & Priority System addresses the competing demands that reach the decision-maker. They solve different parts of the problem.
What leaders should examine before adding more AI
Follow a piece of work from request to completed business action. Look at what execution effort fell, what review or correction appeared, where work waited and whether the final result met the required standard.
Ask the receiving team as well as the person using AI. A quicker handoff is not an improvement if the next team spends longer making it usable. Keep quality and exceptions visible alongside speed.
This is not a recommendation to measure every click or monitor every employee. Select evidence that helps the workflow owner decide what to keep, change or stop. Measure the Workflow, Not the Agent explains how to frame that evaluation without confusing activity with value.
The Leading Space perspective
AI should carry more of the repeatable work around human decisions, not make people the permanent clearing house for an expanding pile of outputs.
Before scaling an AI workflow, ask: Did the complete workflow create more usable capacity? If the answer is unclear, improve the flow of work and decisions before increasing the volume.
Source
Anthropic: How employment is changing at Anthropic, 16 October 2026. The findings above concern Anthropic's own workplace; the business implications are The Leading Space's interpretation.
Turn the finding into a better working system
For one sufficiently bounded company workflow, explore AI Systems Build. Fit and scope are confirmed before implementation.
When the constraint spans workflows, teams, shared information or governance, explore Enterprise AI Systems.