Do agencies still need a PM tool when everyone uses AI?

Yes, but a smaller one, and for different reasons than five years ago.

The honest position sits between two bad answers. "AI makes project management obsolete" is wrong — it confuses coordination with commercial control. "Nothing has changed" is also wrong, because a real slice of what PM tools were sold for is now handled elsewhere.

Worth being precise about which slice.

What AI actually absorbs

Status reporting. Turning a week of activity into a client-readable update was a genuine chunk of account managers' time. Drafting it from notes and commits is now fast. The judgement about what to tell the client is not automated, but the writing is.

Meeting notes into actions. Reliable, and it removes a task people were doing badly anyway.

Chasing and summarising. "What's outstanding on this project" pulled from threads and documents. Useful, and it reduces the number of small synchronisation meetings.

Documentation and handover notes. The thing everyone agreed was important and nobody did. Now cheap enough that it sometimes happens.

Add it up and it's a meaningful reduction in coordination overhead — the layer of work whose only purpose was making sure everyone knew what everyone else was doing. That was a large part of what lightweight PM tools were sold to solve.

So if your tool is mainly a place where tasks are stored and status is typed, its value has genuinely declined.

What it doesn't touch

Everything with money or accountability attached.

Scope boundaries. The record of what was agreed, what changed, when, and who approved it. This is a contractual artefact. A model summarising a thread is not the same thing as a change request someone signed, and when a client disputes an invoice, only one of those helps.

Margin. Cost against value, per project, updated while there's still time to act. This requires accurate time data, loaded cost rates, and a system that holds both. No amount of drafting assistance produces it.

Capacity. Who is available, when, with which skills. This is a commitment you make to clients months ahead. It needs a shared, authoritative view — not an inference from recent activity.

Accountability. A named person, a date, a thing that is either done or not. Someone has to own it, and the owning has to be visible to other people. A summary that says work "appears to be progressing" is not accountability.

The audit trail. When a project goes wrong — and one will — you need to know what was promised and when it changed. Reconstruct that from chat history and see how it feels.

Notice these are the same things that were always underserved by generic task tools. AI hasn't made them less necessary; it's stripped away the layer that was easier to see, leaving the harder layer more exposed.

The shift in what a system is for

Then	Now
A place to record and read status	A place to record commitments and consequences
Coordination: who's doing what	Capacity: what can we sell, and to whom
Task completion as the unit	Scope and margin as the unit
Value from having everything written down	Value from having decisions and money traceable

If the system you're paying for is still mostly delivering the left column, the question isn't whether to keep a PM tool. It's whether to replace it with something aimed at the right column.

Two failure modes to avoid

Cancelling the system and running on chat plus AI. This works for one or two months, and specifically until something is disputed. Then you discover that the record of what was agreed exists as forty messages across three channels, and that the summary you generate from it is a reconstruction, not evidence. Small teams with a single decision-maker can genuinely operate this way. Anyone with fixed-price work and multiple concurrent clients cannot.

Keeping the tool and assuming it's fine. More common. The subscription renews, tasks accumulate, and nobody notices that the questions the business actually needs answered — is this project profitable, can we take on that retainer in October — are being answered in a spreadsheet by one person, badly.

The test

Pick your most complex live project and try to answer four questions without asking anyone:

1. What did we agree to deliver, and has that changed since the proposal?
2. What has it cost so far, against what we quoted?
3. Who is the next action assigned to, and when is it due?
4. If this client asks for another project starting in six weeks, can we say yes?

An AI assistant with access to your chat and documents can probably help with three. It will struggle badly with two and four, because those need structured data — hours against projects, cost rates, forward commitments — that doesn't exist unless something is capturing it deliberately.

If you can't answer two and four, the gap isn't coordination. It's the commercial layer, and that's what to fix.

Where this is heading

The plausible direction isn't the disappearance of project systems. It's a change in who does the typing.

Coordination work — writing the update, chasing the outstanding item, drafting the handover — moves towards being generated rather than authored. The system of record stays, because contracts, invoices and disputes require a record that someone stands behind. What changes is that its job is less "the place people write things down" and more "the place where commitments and money are held."

That makes structured commercial data more valuable, not less. If your time entries are guesses and your rate card lives in someone's head, better tooling of any kind — AI included — produces confident answers built on bad inputs.

What to do

Run the four-question test this week on one real project. Whatever you can't answer is your actual requirement, and it's almost certainly a narrower one than a full platform migration.

Then fix the inputs before the tools. Accurate same-day time entry and a real loaded cost rate do more for your ability to answer questions two and four than any software decision you could make this quarter.