AI can produce useful drafts and confident mistakes. Decide how your team will check its answers before using it for recurring work.
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.
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.
The study found that results depended on whether the task suited the tool.
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:
- which tasks AI is good for
- which tasks require review
- which data should never be pasted into a tool
- which outputs need source checks
- who owns the final answer
- how the workflow changes after AI enters it
Without those rules, employees have to guess what they may share and which answers need checking.
Where AI is usually strong
AI is often useful for:
- first drafts
- summaries
- checklists
- brainstorming
- extracting action items
- rewriting for clarity
- creating examples
- turning notes into structure
- finding patterns in text
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:
- final decisions
- sensitive client communication
- legal or compliance language
- financial recommendations
- medical or safety advice
- complex analysis where the facts are incomplete
- anything that requires deep business context
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
A polished answer can still contain wrong facts or weak reasoning.
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.
Check the answer against its sources even when the writing looks finished.
The fix: review rules
Every business using AI needs review rules. For example:
- AI may draft client follow-ups, but a human reviews before sending.
- AI may summarize a meeting, but the meeting owner confirms decisions and tasks.
- AI may answer internal questions only from approved documents and must cite the source.
- AI may create a report draft, but numbers are checked against the source system.
- AI may suggest next steps, but the account owner decides.
Write the review rules into the task instructions.
The fix: workflow boundaries
Give AI a specific task with clear source material.
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."
The second instruction names the source, the output, and who reviews it.
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.
Measure time saved, errors caught, revenue recovered, or cycle time. Excitement isn't a business outcome.
Look for:
- fewer missed steps
- faster turnaround
- cleaner handoffs
- less repeated work
- fewer interruptions
- better client experience
- hours actually recovered
Keep a record of the results so you can tell whether the setup helps.
Before your team starts
Choose a task and agree how its output will be checked.
Name the person responsible for reviewing the answer and correcting mistakes.