When AI speeds up building, deciding becomes the bottleneck

Projects estimated at eight months shipped in three. My team logged about 30% more effective working hours. Sales growth is the goal we have not reached.

I don’t read that as a failure of AI or of anyone’s effort. I can’t name the company where my team works, because some of what follows is discussed in private, but I suspect it describes more companies than mine.

1. New way of working
2. More things to build
3. More things to decide

01 / WHAT CHANGED

From doing the work to making it repeatable.

At the start of the year the company gave us AI tools and asked us to use them in most of our daily routine.

This is the view of an external technical team inside one company, one that also lets us take part in some of the decisions that involve us. Getting a skill to run without us takes repeated adjustment.

02 / WHAT WORKED

Building got faster. Trying ideas got cheaper.

• PLANNED8
• SHIPPED3

PROJECTS WE HAD ESTIMATED AT EIGHT MONTHS

Hunches that used to be too expensive can now be validated. Our time used to go only to projects with a clear commercial value, because there was never enough of it, and testing an idea was a luxury. Now it’s cheap enough to do.

But every validated idea creates a project that needs follow-up, and follow-up means meetings.

03 / WHAT IT COST

Less typing. More responsibility.

Performance metrics adapted to the use of AI as well. The expectation is now more deliveries in less time, so a faster baseline became a higher delivery expectation, not more slack.

The work did not disappear. It moved into decisions, and that is heavier on the head. It is a price of the new way of working, and one worth naming.

04 / THE CONSTRAINT THAT SHOWED UP NEXT

The bottleneck moved.

The company’s culture gives teams a good amount of autonomy to take initiatives that help the business. Before AI, that worked well, partly because building was slow and expensive: every project had to earn its place, so we prioritized hard. Once the cost of building dropped, that filter got weaker and initiatives multiplied.

FASTER

• Build
• Validate
• Start another initiative

STILL AT HUMAN SPEED

Choose what matters.
Sequence the work.
Coordinate the teams.

CONCEPTUAL FLOW · NOT A MEASUREMENT OF CAPACITY

More delivery is not the same as sales growth

Many of the initiatives we coordinated with other teams have not produced the results we expected. The company shares its growth metrics with us, so we see our part in its wins and misses. Previous years were good. This year the goal we have not reached is growth in sales, even with a steady stream of important initiatives. What did grow is the effective working hours my team logged.

From my seat, the faster pace exposed a limit in direction. Deciding which of many possible initiatives matters, sequencing them across teams and keeping them coordinated all happen at human speed, and I don’t think anyone was set up for this rate. That is not a criticism of leadership. It would be a hard problem anywhere, and my team is part of the pattern, not outside it.

NEXT / WHAT WE’LL DISCUSS THIS QUARTER

Who owns the new bottleneck?

This quarter we will sit down with the other departments we work with, and I’d put that question to all of us, my team included. Three topics are on the table.

1. 01PrioritizeWhich initiatives matter?
2. 02CoordinateHow do teams stay aligned?
3. 03Capture valueDoes delivery move the goal?

A fourth carries into next year: the hours of work will have to be redefined, together with how we use the new AI tools the organization has adopted. That includes a plain question. When implementation gets faster and the time moves into specs, reviews, testing and meetings, is that work counted in the team’s capacity, or does it come out of people’s hours?

None of this is a question about whether AI is good or bad. The projects that shipped faster are real. I don’t have the answers yet, and I’d rather not pretend to.