← Back to Journal

An Agent for Every Client

Two weeks ago, my vision was a Claude project for every client. A shared workspace where the team could pull context, run queries, collaborate with AI on strategy.

That already feels like a half-measure.

The idea surfaced in a conversation with our head of ecommerce. He was looking at custom client dashboards I'd been building — strategy portals with live data, creative performance, CRM insights — and said something that stopped both of us.

"I'm pretty sure in three months, all we're going to do is make agents."

He might be right. And I'm not sure we're ready for what that means.

The shape of the idea

An agent for every client means exactly what it sounds like. Not a reporting dashboard. Not a standardised process. Not a workflow automation. An agent — one that holds every piece of context about that client, understands their data, knows their history, and can build, analyse, and recommend on demand.

I built the first version for a client in about four hours. A strategy portal — CRM data, creative performance by audience segment, lead projections, media planning. All integrated, all interactive, all custom to exactly what that account needs. Not a template applied to their data. A bespoke tool built from their specific situation.

The old agency model runs on standardisation. Build one system, apply it everywhere, make exceptions grudgingly. The economics demand it — customisation is expensive, so you standardise.

AI inverts this completely. Customisation costs almost nothing. Ecommerce brand or lead-gen account, $50K budget or $5M — the agent just builds what's needed.

What this probably replaces

Here's where it gets uncomfortable. We've spent months building workflow automations, standardised reporting templates, systematised processes. Real investment. Real hours. Some of it nearly finished.

We scrapped one of those N8N workflows this week. It worked fine. It just wasn't the right architecture anymore.

The sunk cost trap is recursive. You build the system, the system works, and then the capability leapfrogs the system. The question stops being "does this work?" and becomes "is this the right architecture for what's now possible?"

A Claude project for every client was the vision fourteen days ago. An agent for every client might be the vision now. Something else will probably replace it in another fortnight. That's not indecision. That's the landscape moving faster than any fixed position can hold.

The questions we haven't answered

We're not at the framework stage yet. Pretending otherwise would be dishonest.

What happens to the team?

This is the one that keeps me up. If an agent holds all the context and builds all the deliverables, what does the account lead do? The honest answer right now: I think they become the person who knows what to ask for. The person with the relationships, the judgement, the taste. The one who says "no, that's not good enough" and "yes, but filter it by these two audiences."

I watched this play out last week in a real meeting. An account lead walked through a client's data in a custom-built portal, spotted a gap between two audience segments that no standardised report would have surfaced, and we rebuilt the view on the spot. The human contribution wasn't the building. It was knowing what mattered. That distinction might be the entire job description going forward — or it might be a comforting story I'm telling myself while the ground shifts.

I'm aware that sounds like rationalisation. Maybe it is.

How do you share agents across people?

Right now, I build agents locally. On my machine. In my terminal. Our head of AI and automation is thinking about this from a shared infrastructure angle — agency-level frameworks stored globally, client-specific contexts nested within. Nobody's solved multi-agent access across teams yet. Not properly.

Does the cost structure work?

I spent $200 in API credits in a single day last week. Four agents running in parallel. The output was extraordinary — multiple client deliverables, dashboards, a complete strategy portal. But scale that across an agency and the numbers get real. It's cheaper than a developer. It's not cheaper than nothing.

Is this even the right frame?

Maybe "an agent for every client" is itself a halfway point that gets leapfrogged by something we can't see from here. Two weeks ago, I couldn't see past Claude projects. Now agents feel obvious. In another month, they might feel as dated as the workflow automations we're already scrapping.

The honest position: we're building toward something we can't fully describe yet. The best we can do is keep building, keep testing, and keep being willing to throw away what we built yesterday.

What about the adoption gap?

Process is outpacing adoption. Our head of operations dropped that line in a Monday meeting and it stopped the room. The tools are already ahead of what the team can absorb. We're proving capability faster than people can change how they work.

The instinct is to slow down and run training sessions. But what we've found is that training doesn't transfer — building does. Every person who's crossed over did it the same way: they sat down, built something real for a real client, and couldn't go back. You can't teach that in a slide deck. Which means adoption is a one-at-a-time problem at a moment when we need it to be an everyone-at-once solution. That's a leadership problem, not a technology one. And we don't have an answer for it yet.

Where we actually are

Five people in the company can do this work today. If we can get that to fifteen within a month, everything changes. But "everything changes" is a phrase people use when they can't describe the specific thing that's about to happen.

What I can say with confidence: the first time you build something custom for a client during a meeting — not after it, during it — and they see their data, their creative, their strategy materialise in real time, there's no going back. Not for them, and not for you.

What I can't say with confidence: whether "an agent for every client" is the destination, or just the clearest thing we can see from where we're standing right now.

We're building anyway. That might be all you can do when the ground won't stop moving.