One Concept, One Example, One Question

The concept

AI can now create candidate work faster than many teams can sensibly review it. That is useful when the task is clear and the result can be checked cheaply, but it becomes a problem when the output only looks finished.

I think of this as review debt: the hidden burden created when plausible work arrives faster than it can be checked, corrected, integrated, and owned.

The debt is easy to miss because the first visible result looks like progress. Something exists that did not exist before:

• a draft;
• a code change;
• a summary;
• a plan;
• or a set of recommendations.

But if the work still needs a human to understand the assumptions, detect subtle errors, check the implications, and decide whether it should be used, then the hard part may not have gone away. It may simply have moved.

That means the value of AI depends heavily on the review path around it. A fast draft is useful when the checking work is realistic; it is much less useful when it creates a polished handoff that someone else has to unpick later.

The example

Imagine someone uses AI to produce a long first draft of a senior decision brief. The result is fluent, structured, and much faster than writing from scratch.

That may be a good use of AI if the person who requested the brief has time, context, and expertise to check it properly. They can compare the draft against the source material, challenge the framing, remove unsupported claims, name uncertainty, and decide what should actually go to the senior audience.

But if the draft is forwarded because it sounds plausible, the work has not really been saved. The review burden has moved to the recipient, who now has to work out whether the summary is faithful, whether the right trade-offs have been represented, and whether the recommendations are grounded in reality.

In that case, AI has not reduced the total burden of the workflow; it has moved work out of the first handoff and into a later, more fragile review.

The question

Ask yourself:

Can this be verified cheaply enough that delegation still makes sense?

If the answer is yes, AI may be genuinely reducing the work. If the answer is no, the useful move may be to narrow the task, add checkpoints, improve the source material, or keep AI in an assistive role rather than treating the output as ready to move downstream.

Related

• I wrote about the wider version of this argument in The More Work AI Does, the More Judgment Matters, which looks at why cheaper AI output makes human judgment, review, and accountability more important rather than less.

Elsewhere this week…

• I write about how AI changes the role of data engineers. AI makes the role more important rather than less, because cheaper generation increases the value of architecture, validation, governance, orchestration, and judgement.

  What do you believe about the future role of data engineers in an AI-assisted organisation?

• The SecondBrain Framework update adds a new document-resolution skill for working through open questions and pending decisions in documents. It also improves DOCX table handling for Markdown-based documents, alongside some smaller DTM and wiki workflow improvements from using the system in practice.

Support

Everything I publish here is free to read, and always will be.

I write, research, build, and maintain this newsletter and my website in my own time. If something here has been useful to you, and you would like to help cover the costs behind that work, you can support me through Ko-fi.