Why some people get more from AI
Two colleagues use the same AI assistant, whether that's ChatGPT, Gemini, Claude or Copilot. One finds it 'a bit generic'. The other uses it every day for real work. The tool is the same, so the difference is in how each person works with it. Three habits account for most of the gap:
- Context: you tell the AI who the work is for, what it's for and what 'good' looks like.
- Iteration: you treat the first answer as a draft and improve it over follow-up rounds.
- Judgement: you check the output before you use it, and you decide what to keep.
Think of a capable new temp on their first day. If you hand them a one-line note, you get a guess. If you brief them, then give feedback and check their work, you get something you can use. AI works the same way.
Before and after: one task, three habits
Before (one line):
Write an email about the project delay.
The result is usually polite and vague. It is no use to you without heavy rewriting.
After (the same task with the three habits):
- Add context: "I manage a website redesign for a client. Launch moves from 3 March to 17 March because their content arrived late. Write a short email to their marketing director. Keep it calm and factual, and don't blame them. Offer two options for catching up."
- Iterate: "Good. Make it shorter, put the new date in the first sentence, and soften the second paragraph."
- Review: "List any facts or promises in this email that I should confirm before sending." Then you check those points yourself and edit what's wrong.
Do this now: look at the three steps above and label each one: context, iteration or judgement. Lesson 2 covers how to build richer context. Here you only need to recognise the habits.
Reading AI productivity claims: evidence, hype and fear
Headlines say AI will make you twice as fast, or that it makes people worse at their jobs. Strong users ask the same three questions about any figure:
- Who measured it? Name the source.
- When? Give the date. AI tools change fast, so an old result may no longer hold.
- Does other research disagree? Look for studies or forecasts that point the other way.
Here is how this works on three well-known studies:
- Customer support (Brynjolfsson, Li and Raymond): first an NBER working paper in April 2023, then published in the Quarterly Journal of Economics in 2025. Among 5,179 support agents, an AI assistant raised issues resolved per hour by about 14% on average and 34% for novice and low-skilled workers. It had minimal effect on the most experienced agents.
- Consultants (Dell'Acqua and colleagues, Harvard Business School with BCG, 2023): 758 BCG consultants used GPT-4. On tasks within the AI's capabilities, they completed 12.2% more tasks, worked 25.1% faster and produced work rated about 40% higher in quality. On a task chosen to be outside those capabilities, AI users were 19 percentage points less likely to reach the correct answer. The authors call this a 'jagged frontier': AI helps a lot on some tasks and hurts on others that look similar.
- Experienced developers (METR, published July 2025): in a randomised trial run February to June 2025, 16 experienced open-source developers took 19% longer when AI tools were allowed. They had expected AI to speed them up by 24%, and afterwards they still believed it had made them 20% faster. In February 2026, METR said a follow-up experiment with newer tools gave an unreliable signal, partly because many developers would no longer agree to work without AI.
The lesson is not 'AI works' or 'AI fails'. Results depend on the task, the person, the tool and the date. People's own impressions of speed can also be wrong.
Task 1: Check one AI productivity claim
- Find a claim about AI and productivity in a news article, a vendor page or a post at work.
- Trace it back to the original study or report, not an article quoting it.
- Write down the source, the date of the data, who was studied and what was measured.
- Search for at least one study or forecast that disagrees.
- Rewrite the claim in one honest sentence, for example: "In a 2023 study of support agents, novices gained most."
Task 2: Record your baseline
You will use this baseline to measure your progress through the course.
- Choose three real tasks from your own work that you already use AI for, or could.
- For each task, paste the exact prompt you would use today.
- Rate how satisfied you are with the usual result, from 1 to 5, and write one line on why.
- Note roughly how long the task takes you now.
- Save this somewhere you'll find it again. You'll repeat the same tasks later.
Why nothing is guaranteed
No technique and no published figure guarantees a result for you. Here's why:
- Studies measure specific people, tasks and tools at a specific time. Your work is different.
- Averages hide wide variation. Some people gain a lot and some lose time.
- Tools change. Last year's result may not apply this year.
- How fast you feel you're working can differ from how fast you actually are.
So describe the benefits as possibilities. Say "this may save time on first drafts; I'll measure it", not "this will make me 40% faster".
Self-check
- Can I name the three habits and point to an example of each?
- For one AI statistic, can I give its source, its date and a study that disagrees?
- Have I saved my three baseline tasks, prompts and satisfaction ratings?
Key takeaways
- Strong AI users give context, iterate and apply judgement.
- Treat the first answer as a draft and check it before you use it.
- For any AI statistic, ask who measured it, when, and who disagrees.
- Results vary by task: AI can help on one task and hurt on a similar one.
- Record a baseline now so you can measure your progress honestly later.
- Describe AI benefits as possibilities to test, never as promises.