The Open-Book Trap: When AI Adds Work to Decisions We Already Know How to Make

AI makes information easier to access and synthesize. Turning that advantage into productivity requires judgment about what to delegate, what to verify, and when we already know enough to act.

Before AI, researching a business question might mean opening ten websites, comparing reports, and assembling the useful pieces yourself. Now, much of that work can happen in a single conversation.

That is a meaningful improvement.

But the convenience can obscure an important distinction: easier access to information does not automatically improve the information itself. And faster analysis does not automatically produce a faster—or better—decision.

Consider the everyday choice you already know how to make. You have relevant experience. The consequences are modest. You can explain your reasoning.

Then you think: Let me just ask AI.

Sometimes, that check reveals something useful. Sometimes, it produces three alternatives, two caveats, and another question to resolve before you can get back to work.

The opportunity for leaders is to recognize the difference.

Better access still requires better questions

In many everyday uses, AI draws on information that already existed: documents, research, policies, records, and patterns learned from data. It can combine that material in useful ways and surface connections we might miss. Its capabilities extend beyond retrieving search results.

Still, the underlying material does not become more trustworthy simply because it arrives in a clear paragraph.

Garbage in, garbage out remains relevant.

Imagine asking AI to recommend which project should receive funding. You provide projected revenue, delivery costs, and timelines. You leave out a critical dependency and the fact that the team responsible is already overcommitted.

The recommendation may be coherent. It may even be the best answer to the question you supplied. It can still be a poor decision for your organization.

AI can also introduce errors of its own. Better inputs help, but they do not guarantee a correct output.

Before relying on a recommendation, establish what the system was given, what is missing, and which claims need checking. A polished answer cannot compensate for an incomplete understanding of the problem.

The open-book exam problem

Imagine taking an open-book exam.

You reach a question you understand. You know the answer and could explain it. But the book is available, so you look it up.

You find the answer you expected. You have gained reassurance and spent additional time.

That may be a sensible trade when the question carries substantial weight or your recollection is uncertain. Apply the same habit to every familiar question, however, and the exam takes longer without necessarily becoming more accurate.

AI makes that extra check exceptionally convenient.

An experienced manager knows a routine update needs a clear request and a deadline. Instead of sending it, they ask AI for five versions, compare the tone, request a warmer option, and reconsider the original.

The task has expanded from communicating a decision to evaluating alternatives.

This is an illustration of a possible habit, not evidence that AI universally slows people down. AI can save substantial time. A randomized workplace study involving roughly 6,000 workers found reductions in email time among workers given access to integrated generative AI tools. The benefits varied across activities. Read the research on shifting work patterns with generative AI.

The useful question is specific: What did this additional interaction improve?

If it caught an error, resolved uncertainty, or made the message meaningfully clearer, it earned its place. If it only delayed a sound, reversible choice, the productivity benefit is harder to defend.

Decision fatigue—and the cost of another check

Repeated decisions can feel exhausting. Choosing, comparing, reconsidering, and monitoring all demand attention.

Decision fatigue is a legitimate concern, although the research is more mixed than popular accounts suggest. A 2025 systematic review in healthcare found evidence supporting decision fatigue in some tests, alongside substantial inconsistent and inconclusive findings. Its definitions and measurement remain unsettled. That evidence should not be stretched into a claim that every additional workplace decision depletes a fixed mental battery. Read the review.

We can nevertheless examine the work our AI habits create.

A tool may remove the effort of drafting while leaving us with more versions to evaluate. It may resolve one uncertainty while introducing several possibilities we had not planned to consider.

There is also a plausible risk to self-trust: if we routinely seek external reassurance before acting on familiar, low-stakes judgments, we may become less comfortable making those judgments independently.

That is a hypothesis worth examining, not an established outcome for everyone who uses AI.

A 2025 survey of 319 knowledge workers found that higher confidence in generative AI was associated with less reported critical thinking, while higher confidence in their own abilities was associated with more. The study relied on self-reports and did not establish that AI causes people to lose confidence over time. Read the study on AI, confidence, and critical thinking.

The leadership implication is practical: preserve opportunities to form a view, explain it, test it against evidence, and learn from the outcome.

Confidence should be calibrated through feedback. Neither personal certainty nor an AI response deserves automatic acceptance.

When more analysis becomes Decision Debt

Suppose a leadership team has enough information to choose between two viable projects.

One offers a stronger near-term return. The other strengthens a capability the organization will need later. AI has helped clarify both options.

The unresolved issue is which priority matters more.

Another comparison may be useful if a material fact is missing. But if the team keeps generating analysis because no one wants to own the trade-off, information gathering has become a way to postpone commitment.

This is where my Decision Debt model becomes useful: decisions that are delayed, avoided, outsourced, or surrendered can create costs that compound.

AI can help reduce that debt by making the relevant evidence easier to assess. It can also make postponement easier to sustain.

Before requesting another analysis, ask:

What specific finding could change this decision?

If you can name it, investigate it. If you cannot, clarify who owns the choice and when it will be made.

Give AI a defined role

A useful working approach is to distinguish four responsibilities: automate, augment, verify, and retain human ownership.

These can overlap. Verification supports both automation and augmentation. Human accountability remains even when a system acts automatically.

For a familiar, low-stakes choice within your competence, you may need no AI involvement at all.

For a consequential choice, confidence alone is insufficient. Define the evidence and review the decision requires.

The aim is to match effort to uncertainty and consequences.

Set a stopping rule before opening the tool

Without a stopping rule, almost any question can support another prompt.

Before using AI, complete this sentence:

I am using AI to resolve ______, and I will have enough to act when ______.

For example:

I am using AI to compare these proposals against our three agreed criteria. I will have enough to decide once the cost assumptions and the delivery dependency are confirmed.

That gives the work a purpose and an endpoint.

It also fits the Decisive EDGE operating loop: Context → Decision → Action → Takeaway. Understand the situation, make an explicit choice, act, and capture what the result teaches you.

AI can contribute throughout that loop. Leaders remain responsible for moving through it.

Track whether the process helps: time to a defensible decision, errors or rework, and outcomes worth learning from. A shorter process that produces expensive mistakes is no improvement. Neither is an elaborate process that leaves everyone waiting.

Use the time saved to lead

The promise of AI is substantial: less effort gathering information, more capacity to understand it, and better support for difficult work.

Realizing that promise requires us to be deliberate about the habits we build around it.

Let AI handle appropriate routine work. Ask it to challenge assumptions when that challenge matters. Verify the claims that carry consequences. Keep responsibility visible.

And when you already have sufficient evidence for a familiar, reversible choice, allow yourself to make it.

An open book is useful. Knowing when to close it is part of judgment.