Why is my engineering firm still slow after AI made the work faster?
You sped up the pod and left the org chart alone. The old roles fill the freed hours and rebuild the slow system around your new tools.

Because the tools got faster and the org chart did not. Speeding up the pod, meaning the small team that carries a project, is a change to how the work moves. It is not a change to which roles exist, how many hours each one holds, or what the firm rewards. Until those change too, the structure you built for the slow system quietly takes back the hours the new tools handed you.
What are the three phases of the transition?
I think about an AI transition in an engineering firm in three phases. Most firms run the first two and call it done.
- Phase 1: the tools. Drafting, first-pass analysis, routine checking. You buy or build the tools and an individual task gets faster.
- Phase 2: the work. How a project actually moves from kickoff to submittal. The pod gets rebuilt around the tools: fewer handoffs, less rework, shorter loops.
- Phase 3: the roles. Which jobs exist in the firm, how many hours each one holds, and whether each still earns its place now that the system runs at a different speed.
Phases 1 and 2 are where the speed comes from. Phase 3 is where the speed either survives or gets spent.
Why does the saved time disappear?
In a slow system, a lot of people exist to compensate for the slowness. Someone chases status. Someone assembles the picture so a manager can look at it. Someone relays a comment from the reviewer to the designer and back. Someone keeps the log, the tracker, the sheet that says where everything stands.
None of that work shrinks when the pod gets faster. It does the opposite. More happens per day, so there is more to gather. More revisions move, so there is more to route. The status report is still due Friday, and it still eats a day, because nobody changed the report.
So the hours come back. They come back as coordination around a faster pod, and the firm ends up slower than the tools allow, with more people than the work needs, doing work the client never sees.
Here is the shape of it. If a report that used to take a week of someone's calendar now takes a morning, you freed four days. If the role built around that report still holds five days of coordinating, reviewing and distributing it, you freed nothing. The four days went into the machine that makes the report slow.
Which roles exist only to manage slowness?
Start with what someone produces, not what they know or how long they have been here. Three questions do most of the work.
- What would stop by Friday if this role vanished on Monday? If the honest answer is the status report, the tracker or the meeting, the role manages slowness. If the answer is a design, a submittal, a permit application or a signed decision, it does not.
- Does the role produce the work or a message about the work? Relaying information is a real job in a slow system. It is not a job in a fast one.
- Would a fast pod still need these hours? Not the person, the hours. A role can survive and still shrink from full time to a few hours a week, and that is a far easier conversation than the other one.
Is this a layoff in disguise?
No, and if it turns into one you did it wrong. Phase 3 means retiring roles that exist to manage slowness. It does not mean cutting engineers, and it is not a headcount target dressed up as strategy.
The distinction matters because those roles are usually held by decent people who are good at a job the firm no longer needs done at that volume. An engineer with a license and judgment is not the problem. The problem is a structure where three people touch a document so one person can decide something.
What I would not do is announce a phase and let people guess. Decide what the freed hours buy, then have the individual conversation, early and straight.
Why is the third phase the hard one?
Phases 1 and 2 are technical and you can prove them. Phase 1 you can buy and measure. Phase 2 you can test on one project and compare the schedule. Phase 3 touches titles, pay, identity and the shape of the org chart, and none of that fits in a pilot.
It also collides with how a lot of firms grew. When the system was slow, the way to deliver more was to add people. So a partner's standing came to be measured partly by how many people report to them. Retire the slowness roles and that measure moves, and the person who liked that measure pushes back.
I will be straight about where I am on this. I have run the first two phases in my own firm and I know the pattern of hours coming back. The third phase is the part I have the least tested proof on, so treat what is above as my view, not a finished playbook. The logic is sound. The proof is still being built.
What do I do next week?
Six things, in order. Do not skip ahead to the last one.
- Pick one project where the pod already got faster. Write down honestly what a week of it used to cost in hours and what it costs now.
- List everyone who touched that project in the last month, and write what each one produced that a client or an agency would recognize.
- Mark each name: produced the work, or moved information about the work.
- Take the second group and ask, for each role, what stops by Friday if it disappears. Keep the answers on paper.
- Decide what the freed hours buy before you touch a role. Another project, a shorter schedule, training, a Friday that ends at five. Name it out loud so the firm hears the point.
- Have the conversation with the person in plain words before the org chart changes. If the answer is that the role should shrink to a few hours a week, say so, and give the person a say in how.
The tools gave you the hours back. The only question left is who gets to spend them.
From the book Engineer Your Freedom
Michael Groselle, P.E., is the founder of Modern Engineering Solutions, a multi-state water and wastewater engineering firm. He left salaried engineering in 2017 and built the firm from one resignation letter and zero clients to more than $2M in annual revenue, with a team of 37 and 525+ completed projects.
Sold on Amazon. Prefer the Kindle edition? Buy it on Amazon ($9.99).