Why AI tools slow your team down before they speed it up

You rolled out an AI assistant to save time. A month later, some things take longer. Listing descriptions sit in review. Emails get rewritten twice. Someone spent an afternoon fighting with a prompt.

That isn't a sign you made a mistake. It's the most predictable part of adopting any big new technology.

The dip has a name

Economists Erik Brynjolfsson, Daniel Rock and Chad Syverson call it the productivity J-curve (NBER). Their point is simple. Technologies like AI only pay off once a business also changes how it works: new processes, new skills, new habits. That work takes time and doesn't show up as output. So productivity looks flat, or worse, before it climbs.

Picture the letter J. A small drop first, then a long rise.

What the dip looks like in a small business

Here's how it might go in a 20-person real estate office that starts using AI for listing copy.

• Week 1: Agents try it, get generic text, and rewrite it from scratch. Slower than before.
• Week 3: Two agents find prompts that work. Everyone else has quietly stopped using it.
• Week 6: The owner shares the two prompts that work. Most of the team tries again.
• Week 10: Listings that took 40 minutes take 15.

The dip is the stretch between week 1 and week 6. Most of the time saved depends on how quickly the good habits spread.

Five ways to get through it sooner

1. Pick one or two jobs, not "use AI for everything." When Gallup asked workers what held them back, the top answer was not seeing how AI fit their work (Gallup). "Use it for listing descriptions and first replies to new leads" is a plan. "Use AI more" isn't.

2. Show what good looks like. One real example from your own business beats an hour of general training. Share the prompt, the output, and what the person changed before sending it.

3. Give people time to practice. Learning a new tool on top of a full workload means it gets learned last. Set aside 30 minutes a week, for a few weeks, for trying it on real work.

4. Write down the rules. People hold back when they're unsure what's allowed. One short note answers most of it: what's fine to paste in, what isn't, and when someone must check the output before it goes to a client.

5. Ask every week where people are stuck. The dip lasts longest when nobody knows what's slowing people down. One anonymous question a week tells you, while there's still time to fix it.

Don't judge it too early

The worst time to decide AI "isn't working" is a few weeks in, right at the bottom of the J. Judge it once the habits have had time to form. In the meantime, track the reasons people give you, not just the hours.

If the reasons keep changing ("don't know what to use it for" becomes "the answers need too much checking"), you're moving through the dip. If they stay the same for a month, something needs your attention.

WeeklyTake asks that one question every Monday and shows you the answers on Thursday, so you can see which way your team is heading.