Wasn't AI Supposed to Make Us Faster?

Some weeks my team looks superhuman. Other weeks the same team looks like it’s running behind.

Nothing changes in between. Same people, sometimes the same week. We’ll turn around a quarterly recap in half the time it used to take, with better information in it, better supporting data, better visuals. And in that same stretch, something else on the board is moving slowly, and I can feel the question forming in somebody’s head: weren’t you supposed to be faster now?

For a while I read that as a phase. The story you hear about AI adoption is that teams get slower before they get faster: you pay a learning cost, you ride out the dip, the curve comes back up. It’s a comforting story and it has real research behind it. It also isn’t what this looks like from inside a team.

The dip and the climb aren’t taking turns. They’re happening at the same time, in different places, and that’s a much harder thing to manage than a phase.

TL;DR

• The productivity dip is real, but a team doesn’t live it as a curve. It shows up as variance: brilliant on one thing, behind on another, in the same week.
• Freed hours never bank. They go straight into valuable work you’d never had the capacity to start, which is a real gain that nobody was tracking.
• So the gain is genuinely hard to prove. It lands in a column with no baseline, and there’s no number for anyone outside your team to watch improve.
• Nothing measures this. The only instrument is a team that says out loud what’s actually happening, and a leader close enough to hear it.
• “We’re doing more” is the honest frame today, and it has a shelf life. Eventually they’ll want the existing work done faster, and that’s a different job: automation.
• The white space you promised your team won’t appear on its own. At some point you have to decide to create it.

The curve is real. Nobody lives on it.

The research here is good, and worth knowing before I argue my perspective on the ground.

Erik Brynjolfsson, Daniel Rock, and Chad Syverson described the Productivity J-Curve in 2018: when a genuinely general-purpose technology arrives, measured productivity gets worse before it gets better. Not because the technology is failing, but because the money and hours going into complements — training, new processes, new ways of working — count as cost while producing nothing yet countable. The investment is invisible. The drag is very visible. Then the intangible capital starts paying, and the curve swings up into the shape that gives it its name.

That’s an accurate description of the economy. It’s a bad description of a Tuesday.

Because at the scale of one team, the two halves of that curve don’t happen in sequence. They happen concurrently, in different work. The hours my email strategist spent learning to point an AI at a problem were the down-stroke of the J. The recap report that went out in half the time was the up-stroke. Same month, same team. The scorecard doesn’t read as a curve. It reads as inconsistency, and inconsistency is what people notice.

The dip and the climb aren’t a sequence. They’re two columns on the same scorecard.

The distinction matters because it changes the job. If it’s a phase, your job is patience: absorb the pressure, wait it out, point at the research. If it’s variance, patience does nothing for you. The job becomes attribution, knowing which slowness in front of you is investment and which is trouble, and being willing to say so when someone asks.

Where the hours actually went

Two things happened on my team over the last stretch, and the difference between them is the whole point.

The first: our email journeys need quality assurance before they go anywhere. My email strategist used to do that by hand, every round, and it cost him real time — the tedious, careful kind you can’t rush. This past round he set up an AI agent to run the first pass on the testing. His job became reviewing what it found and going deep on the specifics that deserved it. That’s capacity freed, and a good example of what the work actually becomes: he didn’t stop being the expert, he stopped being the one clicking.

The second: I’ve been sending quarterly recap reports that were never on our list. Not a deliverable we owe anyone, not something we’d ever produced, and genuinely useful, which is exactly why I built them. I send them on behalf of the team rather than from me. That’s capacity spent.

Put those side by side and you can see where the productivity went. It didn’t become slack. It became a longer list of things we deliver, and it will keep going there, for a reason most leaders will recognize.

Like most teams I know, we have close to an unlimited list of things we should be doing and have never had the capacity to start.

That backlog isn’t a queue we’re working down. It’s a sponge. Whatever AI hands back gets absorbed before it can pool anywhere.

Why you can’t prove it

This is where it gets uncomfortable, and I don’t think I’m the only one sitting in it.

If the gains get spent on work that had no baseline, then nothing accrues. There’s no line on a chart that goes from bad to good, because the new work was never on the chart. The recap report can’t have gotten faster; it didn’t exist. So when someone senior asks the fair question — is the AI investment working? — the honest answer is that I’m confident it is and I can’t show you the math.

Every hour AI gave back got spent before anyone counted it.

The reverse is just as hard to prove. Someone asked me recently what would convince me a particular AI bet wasn’t working, and I had a harder time with it than I expected. My instinct when something dips is to look at the person and the pipeline before I look at the tool — is this someone carrying too much, is there a blocker upstream nobody solved? That instinct is usually right, and it’s a useful correction to the reflex of blaming whatever is newest. But held too tightly it becomes unfalsifiable, and a bad bet can hide inside it for a long time.

For now my answer is that we’re in the learning phase, and in the learning phase nearly every attempt pays — if the thing itself fails but the person comes out of it knowing how to direct these tools better, that’s still a return. I believe that. I also know it has an expiration date, and I’d rather name it now than discover it later.

The only instrument is a team that talks

There’s no dashboard for this. Nobody’s status update says “I lost six hours learning the tool.” It says the task took longer.

So the only thing that has actually worked is talking about it openly, constantly, inside the team. We discuss what we’re trying, what worked, what wasted an afternoon. It isn’t a soft practice or a culture nicety. It’s the measurement system, and the only reason I can tell you which slowness was investment.

And it has a hard boundary. Inside my group, everyone knows how a given deliverable really got made. One ring out, nobody does. My leadership has almost certainly deduced that some of what’s landed came from me working with these tools rather than from the team working the old way. The timing isn’t subtle. But we haven’t had that conversation head-on. It’s not a secret anybody’s keeping. It’s a conversation nobody’s started, which is a different and much more common problem.

“We’re doing more” has a shelf life

Right now the frame I use upward is that we’re doing more. It’s true, and provable in the sense that the outputs exist, and so far the response has been appreciation rather than pressure. The bar hasn’t moved.

It will. As we keep exceeding what was expected, expectations follow. Not cynicism, just how it works, and I’d rather plan for it than be surprised. The ask underneath it is already audible: leadership also wants the existing work done faster, not only more of it. “We’re doing more” doesn’t answer that. So the next real push for us is automation, taking work that’s already on the books and shortening it, so the gain finally shows up somewhere a person outside my team can see it.

It’s the move that converts scope back into speed, and I don’t think you get to skip it.

Somebody has to decide where to draw the line

When I wrote about closing the AI skills gap, I said the goal was for these tools to buy the team some slack — room to think, to try things, to get to know how this stuff works. I meant it. I’m also aware that I’ve spent every hour of it.

I keep assuming the white space arrives eventually: enable enough of the workflow, and the room appears on its own. Maybe. The more likely version is that the line gets drawn on purpose or it doesn’t get drawn at all. At some point I have to look at that long list of useful things we could now start, decide which ones are worth doing this quarter, and let the rest wait so the hours survive. That’s a judgment call about strategy, not about tooling, and it’s mine to make.

It connects back to what I wrote about everyone becoming a leader now. When the constraint stops being capacity, the scarce thing becomes deciding what deserves the capacity. Every person on a team with a capable AI at their elbow will face a small version of that choice, most days.

The white space doesn’t show up on its own. At some point, as a leader, you have to decide to create it.