Project management is often treated as the ugly stepchild of agency departments. Clients don’t like paying for it, agencies never know if it lives in delivery, client care, or operations, and being able to fund it as a standalone function is often a puzzlework of margin math.

Ask any agency owner who's had to defend project management as a line item on an invoice to a client or, worse, to their own P&L. It gets treated like overhead; a cost you tolerate rather than something you're actually buying. But somebody is always doing the work of project management, whether or not you've priced it into the SOW.

A fundamental truth of agency life: Project management happens, whether you have a specific project manager doing it or not. You need to account for it.

At an agency, PM's real job (beyond keeping clients and staff sane) is solving what we've called The Problem of Context. Clients hire you for two reasons. One is the business outcome they've judged matches your domain of expertise. The other is that you can be the keeper of context around that work so they don't have to worry about what's lost when an employee leaves, gets promoted, or gets fired.

But a huge issue within many agencies is how you get that context - the thing sales knew, the thing the client said in kickoff, the thing delivery figured out three weeks into the project - moving between people and departments without losing fidelity.

That's not a nice-to-have. It's part of the actual product you are delivering. Clients (for the most part) won’t hire an agency for a single person's expertise; they hire because they want an outcome that doesn't disappear the moment one person takes a vacation, gets promoted, or leaves.

That's always been true. It's why the sales-to-services handoff is the moment that either sets a project up to succeed dooms it. And it's why a client who trusts your agency as a team — not as one favorite consultant — is a client you can staff flexibly without them flinching.

What AI Can Absorb

AI is genuinely good at the mechanical layer of this problem. Recaps. Status synthesis across six threads and three meetings. Flagging that a client's tone shifted in the last two emails. Tracking blockers so nobody has to hold them in their head. If your agency can solve for how to get meeting transcripts, emails, recaps, and status updates from the team into an LLM, you can automate things that used to eat up a PM’s entire Friday afternoon.

What AI doesn't do is judgment. It can tell you the client mentioned budget concerns twice this month. It can't tell you whether that means the project's actually at risk or whether that client just talks that way (though it will absolutely pretend it is providing judgment on that). It can summarize the SOW. It can't tell you when the SOW promised a “full” lead scoring build, but the actual implementation going on is something subtly more narrow. 

So project management, like many roles that were primarily administrative or synthesis-based, isn’t disappearing per se. It's evolving: away from moving cards between kanban columns, and toward defining what those columns should be in the first place; toward reviewing AI output for accuracy, building the structure the agency's automation runs on, and feeding what's learned back into that system. The role starts to look less like a coordinator and more like a product manager for the deliverables PMs previously produced by hand (recaps, updates, etc.), and for the systems through which the agency’s deliverables get produced.

This is especially acute in MarTech implementation, where the context is buried in platform configs, integration logic, and technical decisions nobody wrote down. Getting this kind of context out of the platform and into an LLM in some structured way can save hundreds of hours a year in meetings between PMs and consultants, or between consultants themselves, and help reduce the onboarding time of new consultants to the project.

Where this Leaves Agencies

Project management was always a force multiplying function. The math used to justify the role always should have been some version of “adding 1 PM per 6 employees allows me to free up 8 billable hours per week within the team, which is how I justify that PM’s salary.” Now that force multiplication may look more like “adding 1 PM per 12 employees allows me to free up 16 billable hours per week within the team because they have reduced the amount of context gathering meetings and activities the team has to attend.” 

As the mechanical work gets absorbed by AI, what's left is the part that was always the actual value: knowing when something's wrong before the client says so, knowing when "the SOW says X" and "what the client needs" have diverged, and knowing the delivery team well enough to catch it and correct it.

Agencies that treat this shift as an excuse to cut the role will lose one of the things clients were actually paying for. Agencies that use it to free their PMs from busywork and instead let them spend that time on thinking through the design of the automation systems will out-margin the ones still pricing PM as overhead.

Gut Check: Are your PMs moving toward automating their busywork and jumping both-feet-in on figuring out how to make their work and the work of the agency more efficient or do you just have people who provide status updates an LLM could provide with the right inputs? How you structure your team and what you can and cannot automate are the exact questions that our Diagnostic is designed to answer. Let’s chat.