Essay · 2026

AI Made Information Cheap. Attention Is Still Expensive.

We can generate research, analysis and documents faster than we can absorb them. The next AI productivity skill may be knowing when to stop.

Reading time
3 minutes
Published through
LinkedIn and Astana Hub
Topic
Systems and Implementation

About this edition. The concise version was published on LinkedIn and the long-form Russian edition on Astana Hub. This page restores the full argument in English.

Almost any question can now become a small research operation. One agent scans the market. Another compares documents. A third identifies risks. Within minutes, a team has a table, a presentation, three scenarios and a detailed report.

We read the opening, scan the conclusions, press Accept and move on. A week later, someone may launch almost the same research because nobody remembers exactly what the first report said or where it was stored.

We have become faster at producing information than at digesting it.

The bottleneck moved

Generative AI reduced the cost of a first draft. Reading, verification, judgement and integration still require human attention.

A report in a shared drive is not yet organisational knowledge. Until someone has examined the assumptions, connected the findings to a decision and understood what changes next, it remains material waiting to be processed.

Every output creates future work

Someone still has to understand the material, verify the parts where error matters, connect it to existing knowledge, act on it, store it intelligibly or delete it and revisit it if later work depends on it.

When these obligations accumulate faster than a team can resolve them, the organisation develops information debt. Folders look full and polished while questions are researched repeatedly and old findings continue to circulate after their assumptions have changed.

Information hygiene begins before generation

Before starting another AI research task, complete one sentence:

After reading this, we need to decide...

Then ask four questions:

  1. Which decision or action should change?
  2. What is the smallest useful output for that purpose?
  3. Who will read it, and what must they verify?
  4. What happens afterwards: retain, update or delete?

If the sentence cannot be completed, the request may not be ready. If one comparison page is enough, a forty-page report adds processing work without improving the decision.

Decide what deserves to survive

The durable record may need only the decision and rationale, the facts and sources it relied on, material limitations, unresolved questions, commitments and owners.

Not every AI conversation belongs in organisational memory. Not every intermediate draft deserves a permanent place in a knowledge base. Expiry, deletion and a deliberate "this is enough" are part of information hygiene.

AI can summarise the report, the summary and the combined brief. It cannot remove the moment in which a person has to understand the meaning, notice a contradiction and accept responsibility for a decision.

Working in public

Analysis is only useful when the next operational question is visible.

I publish field notes to show how I move from a requirement or risk into product behaviour, control, evidence and ownership.

See the advisory approach

Continue reading

AI Is Already in the Workflow. What Changes Now?

AI rarely enters a company as one large programme. It arrives as a browser tab, a plugin or a familiar service with a new capability.

Read site edition

A Designed System Is Not Yet a Working System

A document can be completed in a week. A mechanism becomes real only where someone makes a decision and either passes through it or works around it.

Read site edition

A Human Clicked Approve. Was the Decision Actually Reviewed?

The button records human presence. Control begins where disagreement can change the outcome.

Read site edition