Why the Expert in the Room Still Matters

AI can execute on a clean spec. It cannot optimize a PSA implementation. Here is why experienced consultants still matter, and what using AI well actually looks like.

Recently I was listening to an NPR segment that really pulled me in. They were talking about how AI is now capable of doing the work of paralegals and first and second-year associates at law firms. The repetitive research, document review, the foundational work that junior staff have always owned. And honestly, that's probably true. The technology is there.

The question that stayed with me wasn't whether AI can do that work. It's what happens to the pipeline.

The partners at those firms, the senior attorneys everyone wants on their case, got there by doing exactly that work. The late nights with case files. The hours of research that felt like grunt work at the time. The mistakes that cost them sleep and taught them things no briefing document ever could.

If firms stop building that foundation because AI can handle it, where does the next generation of senior partners come from?

The answer, I think, is not to resist AI. It's to understand what it actually takes to use it well. And that's where expertise becomes more important, not less.

The hard way is what makes the fast way possible

In my early years of consulting, I remember configuring tables, getting it wrong, going back, researching the fix without the search tools or AI that exist now, trying again, and then validating that what I thought would work actually did. That validation piece mattered then and it still matters now. It just takes me a lot less time.

The same pattern held as the work got bigger. Requirements documentation that once took me far longer. Data migration work that felt enormous and is now something I can move through efficiently and with confidence. The time I put in doing those things the hard way is exactly what made me fast, accurate, and thorough doing them now. It looks like a graph trending up and to the right, where the investment of hours early on compounds into expertise over time.

What those years gave me is depth. The ability to look at a situation and know what is wrong with it before I can fully articulate why. That is not something a prompt produces.

That foundation is also what lets me use AI effectively today. I know when the output is right. I know when it's close but missing something critical. I know when to trust it and when to push back. That judgment is not built overnight, and it is not built by skipping the hard work.

What using AI well actually looks like

Here is the practical version, because "a good consultant uses AI" has become the kind of line anyone can say.

I use AI coding tools, Claude Code and Codex among them, to build the scripts and automations that used to sit at the bottom of a backlog because they were never worth the hours. A validation rule that stops bad data at entry. A cleanup routine for a migration that arrived messier than promised. A scheduled script that flags timesheet gaps before they become a month-end problem. Work that used to mean a scoping conversation and a separate budget line now often fits inside the engagement.

I am not a software engineer and I do not present myself as one. What I am is the person who knows exactly what the script needs to do, how the system will behave when you push it in that direction, and what a wrong answer looks like when it comes back.

These tools are very good at producing something that looks right. Deciding whether it is right is a different job.

Getting a useful answer out of a model in the first place is its own skill, too. Most of the disappointing AI output I see traces back to a thin prompt. The model was never given the process, the constraints, the way this particular organization books revenue, the three things that are true here and true almost nowhere else. So I feed it the requirements documentation, the field structure, the integration behavior, the edge cases I already know about because I have hit them before.

The quality of the output tracks almost exactly with the quality of what goes into it. Knowing what to put in is not a prompting trick. It is domain expertise, applied. And knowing when to stop and think instead of prompting again is the other half of it.

I have been in the services and consulting world for over twenty years now, with more than a decade specifically in professional services automation. The way I see it, the consultants who are going to matter in the next decade are the ones who bring that depth of expertise and know how to put AI to work in service of it.

What AI genuinely does well, and where it stops

I want to be fair about what AI can do in a PSA implementation or any software rollout: direct data migrations where the data maps cleanly, building out configuration tables, certain integration setups where the requirements are well defined and documented. Give it a clean spec and a clear path and it can execute.

Most engagements do not look like that, and implementations are only part of the picture.

I was recently involved in an extensive optimization engagement with a client whose PSA setup had been in place for years, since their original implementation. On the surface, the system was working. Underneath, there was a lot that had never been fine-tuned, and it showed. Most clients I work with on optimization land somewhere in the middle of the maturity model I wrote about previously, where the system runs but underperforms in ways nobody yet called out.

That engagement required deep collaboration with the client, a clear understanding of what was actually happening versus what they thought was happening, and the ability to foresee the end result, which comes from years of expertise in both professional services and project management. I understood the issues, made targeted recommendations, and executed on them. That was human expertise, start to finish.

Even in a standard implementation, what looks routine rarely stays that way.

A client comes to me with what sounds like a simple requirement. Once I get into their actual process and understand how their business works, it almost never stays simple. I worked with a client recently where the way they recognize revenue required a completely custom setup. The solution involved a combination of integrated fields, a custom subset of tasks distinguishing project planning from time tracking, and scripting that tied everything together to feed their revenue recognition process correctly. It also met their requirements for how the services team tracked and managed work day to day.

None of that came from a framework. It came from extensive fact-finding, human judgment, and experience with how these systems behave when you push them in a specific direction. AI did real work in that build, the scripting especially, and it moved faster because of it. But the shape of the solution came from knowing which questions to ask and having seen enough variations of similar problems to recognize when something needed a creative answer.

Would AI alone have caught that? No. It would have looked at the surface requirement and suggested the standard path.

AI can help you get set up. It cannot get you optimized.

Every organization I work with has the same high-level goal: they want their PSA tool working for them, not the other way around. But how they get there varies significantly from one organization to the next. That gap between set up and optimized is where experienced consultants, the ones who also know how to use AI well, do their best work.

The part that worries me

On the layoffs across the tech sector: I watch those headlines and I am pragmatic, though not without frustration. Restructuring happens. What gets under my skin is when the specialized talent walks out the door, the people who carry institutional knowledge, long-standing client relationships, and expertise that took years to build. Those are not roles you backfill with a tool. The impact shows up later, when a client relationship goes sideways or a renewal doesn't close the way it should have.

Senior consultants and specialists are not overhead. They are the reason clients come back.

If you are evaluating whether to bring in a consultant for your next PSA implementation or optimization project, the question is not whether AI has a role in the process. It does, and a good consultant will use it. The question is whether you want someone who has spent years in the weeds of these systems, who knows when AI is giving you the right answer and when it's giving you the easy one, and who can apply that judgment to your specific situation.

The experience is what makes the AI useful. And the experience does not come free. But neither does getting it wrong.