Field Note · Applied AI and system ownership
Does This AI Use Case Automate, Enable, or Create Busywork?
I have started sorting AI use cases into three buckets. The third bucket matters because a surprising amount of AI work does not automate or enable anything. It creates output someone else must now review, correct, move, and explain.
A thinking frame by Andrew Moss
The questions I get
Usually some version of these:
- Which AI projects should we fund first?
- Is time saved the right measure of value?
- How do we tell transformation from more generated work?
What a lot of people seem to think
Any task completed faster with AI is treated as an AI win, and enough faster information is assumed to add up to transformation.
How I look at it
Automation matters, but it has a ceiling. Enablement changes what the business can do. Busywork creates plausible output somebody else must review, move, correct, and explain. My additional test is consequential action: if faster information does not remove total work, improve a decision, change responsible action, or expand a valuable capability, the business may simply be moving the bottleneck.
Why the decision matters
The cost is rarely confined to the line item.
If the sequence is wrong
The company celebrates generation volume while review burden, fragmentation, and hidden rework rise.
If the sequence is right
Automation removes real cost, enablement expands valuable capability, and every project has an owner and measurable outcome.
How reversible is it?
Usually moderate project by project, but a portfolio of weak use cases can consume attention and make the organization cynical about AI.
The short answer
Classify the bucket, then test the action.
Bucket one automates work the business already does. Bucket two enables valuable work it could not do consistently before. Bucket three creates a new review queue. Fund the first two only when the user, decision, total work, responsible action, owner, and measurable outcome are clear.
Three buckets plus one testAutomate · enable · busywork — then ask what action changes.
Automation makes an existing path cheaper or faster. Enablement gives the business a different shape. Busywork produces more material to process. The consequential-action test reveals whether any of those outputs reach the operating result.
Move fromAI output and time saved→Move towardChanged capability, decision, or responsible action
The order I would use
Take the right steps in the right order.
- 01
Name the user, decision, and action
Identify who consumes the result, which decision it supports, and what they can responsibly do better because it exists.
- 02
Classify the bucket
Decide whether the project removes existing work, enables new valuable work, or creates a new review queue.
- 03
Measure total work
Count preparation, review, correction, exception handling, integration, and follow-through, not just generation time.
- 04
Test the action delta
Ask what consequential decision, responsible action, customer result, or business capability changes if the project succeeds.
- 05
Test for compounding
Ask whether verified context and confirmed corrections make the tenth use more valuable than the first.
- 06
Assign an outcome owner
Name who owns adoption, quality, operating change, and the result beyond the AI team.
Questions worth answering
Before the next irreversible move:
- What becomes possible that was not possible before?
- Whose work disappears and whose review burden grows?
- What consequential decision or action changes?
- Does verified use make the tenth result better than the first?
- Which business result will move if the project works?
What not to do
Do not call faster information transformation.
Do not measure only model time. Do not automate a broken step that should be removed. Do not create a new stream of summaries, drafts, or dashboards without a user, decision, action, and outcome that need it.
Keep the perspective
The value of intelligence appears in what the business can responsibly do next.
Efficiency competes with the cost of today’s labor. Enablement competes with the business’s ambition. Both must reach an operating result; otherwise the AI program is producing motion rather than leverage.
Independent sources
Useful primary material
These sources support the public frame. They do not replace the private facts or the accountable professional.
Common follow-up questions
Is automation a lower-value use of AI?
Not inherently. Some automation has excellent, dependable economics. The mistake is calling all efficiency work transformation or assuming savings automatically create a new capability.
What makes intelligence compound?
Verified context, reusable learning, correction loops, and repeated decisions can make later uses better. Unchecked memory can compound error instead.