Can AI give people their attention back?

How better context and review can reduce AI rework and protect judgement.

I want AI to give people more room to think. That means reducing avoidable cognitive load, the mental effort of keeping track of information and working through it. I look beyond how quickly AI produces something to what happens when someone else has to use it.

Consider a weekly project update prepared with AI. It accurately explains how decisions have moved since last week. The slides do their job as a record of progress.

But the sponsor's circumstances have changed. They now need an operational result this quarter to support another business priority. The update says little about whether the project's progress helps with that need.

The deck remains factually correct. A change that affects how leadership understands and supports the project has been missed.

Someone experienced now has to connect the progress report to that changed priority. Whether the project gets the right support may depend on that context reaching the workflow before the next decision.

Automation has saved production time while leaving substantial thinking for the recipient. The document still needs work to contribute meaningfully to the current objective.

The workslop research from BetterUp Labs and Stanford Social Media Lab describes how polished AI output can leave colleagues resolving what it failed to address.

I think of that handover as an AI slop grenade. The slides look finished, but the recipient still has to work out what the progress means for the changed business priority. AI has saved production time while passing unfinished thinking to someone else.

In my earlier writing on AI efficiency gains, I argued that released capacity needs an investment plan. Keeping AI useful as circumstances change deserves a place in that plan. So does maintaining the judgement needed to recognise what it has missed.

Capture the change where the work happens

"Give AI more context" sounds straightforward until someone has to maintain it.

I would start where people already review their work. At the end of a project meeting, or while correcting a draft, capture what changed and what it means for the next output.

A short dictated note might be enough:

"The sponsor needs an operational result this quarter. Keep the longer-term proposal, but explain what we can deliver sooner and what we would have to defer."

AI could propose that update from approved meeting notes. Where an experienced colleague recognises an implication the notes never captured, their contribution is to make it explicit, including where it still needs checking.

The project owner should maintain the current position and the reason it changed. Review temporary instructions at the next meeting. Keep firm-wide standards separate, so one stakeholder's preference does not become a rule for everyone.

Saving the update is insufficient. The workflow needs to retrieve it before preparing the next report.

Review the argument before producing the output

AI could first show its proposed argument: lead with the result achievable this quarter, explain the trade-off and retain the longer-term destination. The reviewer can correct the direction before it becomes twenty slides.

It could also compare the argument with the previous update and flag changed assumptions. The person responsible can examine whether those changes matter and whether the evidence supports the recommendation.

Anthropic's guidance on evaluating AI agents distinguishes what a system says it accomplished from the outcome it produced. It evaluates the model together with the tools and instructions around it.

I would extend that assessment through the business handover. Did the report help its audience make the required decision? How much work remained before it could?

Once the argument is clear, we can also reconsider those twenty slides. Google's Audio Overviews can turn source material into a podcast-style briefing. ChatGPT can help create infographics, while Claude can build interactive pages and dashboards. The audience might benefit from exploring the implications themselves or listening before a discussion.

The medium deserves an article of its own. I'll try to resist making it a deck.

Whatever the format, corrections should improve subsequent work. If a reviewer repeatedly adds operational implications to leadership updates, AI could suggest retaining that instruction. The reviewer decides whether it applies generally or only this time.

Keep people practising judgement

Better context does not, by itself, develop the judgement of the person using it.

If people routinely accept recommendations without examining the reasoning, they may lose opportunities to learn how to challenge them.

Anthropic's coding-skills experiment found weaker immediate understanding among participants assigned AI assistance. It was a small study of learning an unfamiliar programming library. It does not establish inevitable decline across professions, but it gives us reason to assess understanding separately from completed output.

I would use selected difficult cases for that practice. Ask the person responsible to explain their recommendation, then change an important assumption and discuss whether their answer still holds. AI can help generate the alternative case. Experienced colleagues can challenge the reasoning.

Released capacity should also let people be more present with clients. Listening carefully is how they discover some of the changing circumstances that no project folder yet contains.

I would test this approach across related workflows. Ask recipients whether outputs were usable as received or needed substantial rework. Use existing review comments to identify recurring problems. Include the effort spent maintaining context when assessing the time saved.

For me, AI enablement should make it easier to keep work relevant as the business changes. The test is whether people can devote more attention to the decisions and relationships that require their judgement.

Time saved is useful. Time saved explaining the time-saving tool is better.

Tushar Singh is Head of AI and Data at ELab AI, which advises organisations on AI enablement and governance across New Zealand, Australia and Hong Kong.