Field notes

AI can make good workers faster. It can also make wrong answers faster.

By Ashley · The SAGE Stack · · 4 min read

AI at workJudgmentRisk
The short version

AI speeds up good work and bad work alike. The research shows real gains on tasks inside AI's capability range, and worse accuracy on tasks that only look like they fit. The businesses that benefit decide task by task where AI helps, where a human reviews, and who owns the final answer.

AI can make good workers faster.

It can also make wrong answers faster.

That's the sentence every business should hold in its head before rolling AI into operations.

The productivity evidence is real. In customer support, software development, writing, and consulting tasks, studies have found meaningful gains from generative AI assistance.

But the caution is just as real. AI doesn't know whether it belongs in every task. It can sound confident when it's wrong. It can produce polished work that hides weak reasoning. It can encourage people to trust an answer because it arrived quickly and looked finished.

AI adoption needs judgment and rules.

The evidence comes with a warning

In a Harvard Business School and Boston Consulting Group study, consultants using GPT-4 completed more tasks, worked faster, and produced higher-quality results on a set of tasks within AI's capability range.

That's the good news.

The same study found that, on a task selected to be outside AI's current capability frontier, people using AI were less likely to produce correct solutions than people without AI.

That's the warning.

AI helped when the task fit. It hurt when the task looked like it fit but didn't.

That's why the researchers use the phrase "jagged frontier." AI capability is uneven. It's good at some things, weak at others, and the boundary isn't always obvious.

The real business risk

Employees are probably already using AI. (Here's why, and what to do about it.)

The risk is that no one has decided:

Without those rules, AI becomes a guessing layer inside the business.

That creates scattered experiments, not a strategy.

Where AI is usually strong

AI is often useful for:

These tasks still need review, but they're good candidates because the human can usually check the output quickly.

Where AI needs more caution

Be careful with:

AI can support these tasks. It shouldn't own them.

The more the cost of being wrong rises, the more human judgment needs to stay visible.

The problem with polished output

Messy work often announces itself.

Polished wrong work is more dangerous.

AI can produce a clean paragraph, a confident recommendation, or a tidy summary that feels finished. That finish can lower people's guard.

This is especially risky when a team is busy. A polished draft saves time, so people may skip the hard part: checking whether it's true, complete, and appropriate.

That's how AI creates "work-looking work."

The output exists. The thinking may not.

The fix: review rules

Every business using AI needs review rules.

For example:

These rules protect the judgment while still letting the business get the benefit.

The fix: workflow boundaries

AI works better when it has a job description.

Bad instruction:

"Use AI to be more productive."

Better instruction:

"Use AI to draft weekly client updates from approved project notes. The project owner reviews before sending."

That second instruction has a workflow, source, output, and review rule.

That's how AI becomes operational instead of vague.

The fix: measure the result

If AI is supposed to save time, measure time.

If AI is supposed to improve follow-up, measure follow-up.

If AI is supposed to reduce admin, measure admin.

Don't settle for "people seem excited." Excitement isn't a business outcome.

Look for:

That's the difference between AI theater and AI operations.

The bottom line

AI is powerful.

That's why it needs structure.

The businesses that benefit will know where AI belongs, where it needs a human, and how people stay accountable for the work.

The next step

Get the hours back.

The SAGE Stack helps small businesses build AI into workflows with clear boundaries, review rules, and measurable outcomes. The 10-Hour Map finds where AI can safely give time back first.

Sources