Will AI replace your agency, and how do you plan without knowing?

AI will absorb parts of what your agency does, and no honest person can tell you which parts will still be yours a year from now. So the move is a short loop instead of a long plan: work out what the market needs today, sell that, then check again in three to six months and change when the answer changes. You are an entrepreneur first and an agency owner second.

What does it mean when the platform ships the thing you sell?

Start with the part that is settled, because the vague version of this conversation has run for two years and the specific version only started recently.

In July 2026 Klaviyo published a product page for Composer, an AI agent built into the platform itself. The page lays the workflow out in five steps. Confirm your brand guidelines. Describe your goal in plain language. Review what comes back, which is a fully designed email with a subject line, on-brand copy and a suggested audience. Refine it in conversation. Approve and launch. The same page notes a standalone version that works with third-party platforms, and grounds the output in what it calls fourteen-plus years of marketing intelligence drawn from billions of consumer interactions across 193,000 brands.

Read those five steps again and notice what they are. They are an agency's scope of work, written out as a product feature, by the company whose invoice the client is already paying.

You cannot argue with that and you should not try. AI is going to commoditize some services and productize work that used to sit inside an agency. The email writing process, as a billable line item, is going away. That is the ground you are standing on, and nobody gets to opt out of it.

Is a shorter engagement the right response?

One agency owner posted his version of this in September. Six months of flat revenue that did not feel like a sales problem, three clients gone in a month against years of strong retention, and the thing bothering him was that campaign building had quietly become a feature of the software his clients already bought. His proposal: ninety days at a fixed fee, build the flows in the client's own account, spend months two and three training whoever runs it afterwards, and on day ninety he is out. His own doubt, which he put in writing, was whether he had invented a real offer or talked himself into a staffing agency with extra steps.

I think he is thinking well, and most of the replies he got missed it. He has looked at where the market is going and started moving toward it, giving up a service line he could have spent another two years defending. That instinct is correct and it is rarer than it sounds.

My argument with him is about the clock and the ownership, not the idea.

Ninety days is too short. That is not enough time to bring somebody new on at the client and get them genuinely capable, because somebody still has to own it afterwards. Somebody still has to sit in front of the platform and tell the agent what to do, and that job is real work that needs a trained human. It is also a low-paid role. You should not be paying consultant rates for a person who instructs an AI agent, and a client who works that out six weeks in will resent the invoice.

A year makes more sense. Frame it honestly: we are transitioning from my team doing the service to me coaching your team to do it, and here is what the twelve months looks like. That is a different sale from a ninety-day exit, and it is a more truthful one.

The other half of the work is unglamorous and nobody talks about it. That owner needs to learn every single feature in the ESP that is going to replace something his agency currently does by hand, and then work out how to teach it better than the vendor's help docs do. That is months of study with nothing billable attached to it.

Why won't anyone tell you what to build instead?

Because they cannot, and most of the people answering this question are pretending otherwise.

I have been saying this to clients lately and I will say it here. Anyone who tells you confidently what the next twelve months look like does not know. They may sincerely believe they know. I cannot see how anybody could. If the AI bubble bursts tomorrow and we hit a real recession, a great deal of this gets paused. If the grassroots movements against data centres gather momentum, that changes things too. Every week the news carries another advance nobody had priced in. Under those conditions, a twelve-month prediction is a guess wearing a suit.

That is genuinely uncomfortable to write, because people love being told where we are going, and the pages competing for this question all offer a destination. I would rather be useful than comforting.

The deeper mistake sits underneath the prediction anyway. When you pick a destination and march at it, you are superimposing your idea onto the market instead of letting the market tell you what it needs. That gets neglected constantly and it is the whole ballgame right now. The most agile companies look at the changing ground and say: I have no idea where this is in a year, but I can see what the market needs this quarter. So they market that and they sell that. Then in three or six months they ask whether it is still what the market needs, and they adjust, optimise, and change again, because the need is moving. It moves on a lag behind the technology, and it moves fast.

An entrepreneur cannot get attached to a particular idea, or a particular way of doing things, and stay in business through a period like this one. Entrepreneur first. Agency owner second.

What does reading the market actually look like?

You listen, you watch, and you use your brain. That is the whole method, and it costs you the thing you are least willing to give up.

It means being pulled out of the execution of the work and into thinking for your clients. There is still a place for execution. There has to be enough time carved out to anticipate, though, and most owners have none, because the billable work expands to fill the week and thinking feels like the thing you do once the real work is done.

The richest signal is on sales calls, and it is not subtle once you know what you are listening for. Prospects will tell you. It comes out as do you guys use AI. It comes out as I read this interesting article. Most business owners believe they have a finger on the pulse of what is happening in tech, and some of them do, but they are not saturated in it the way consultants and agency owners are. We sit in this all day. They read three pieces on a plane.

And then the sharp ones, which are the ones worth paying attention to: do we even need you anymore? and do you think we could bring this in-house?

Any news about AI should start a conversation about AI, because that conversation is becoming more critical to the sales process every month.

I am in a long sales conversation right now with someone who has said they would like to hire me. Since we started talking, AI has moved enough that the shape of the offer may need to change before we finish, and they have raised it themselves. Treat that as market research delivered to your door, about an offer you were about to sell, by the person who would have bought it.

When the harder version lands, do not reach for a script. The prospect is pressure-testing you, and they should be. They want to watch you think, and being pushed onto your heels a little is part of the exchange. What I come back to is the fact that they are on the call at all. There is always a reason. So I ask them early: you are a busy owner and you are giving me thirty minutes today, which means you see some value in this conversation, so tell me what you think is valuable about the meeting we are about to have. If the answer is that they were just honouring the time, you have learned something useful very quickly and you can both stop.

What does it cost to actually change?

More than the people selling the change will tell you, and this is the part I watched up close growing up.

My dad farmed, and he was never afraid of a better method. When we moved from broadcasting fertiliser to anhydrous ammonia, the change was obviously right. Broadcasting means spreading it across the top of the field and waiting for rain to leach it down. Anhydrous means injecting it straight into the ground with a cultivator, which is more efficient, better economics, and kinder to the land. Easy decision.

And then the bill for the decision arrived. Cultivators that had never carried hoses had to be outfitted. The product comes out of the tank so cold it freezes anything it touches instantly, so there was a whole safe-handling procedure to learn. You either arranged delivery or you got trained to tow pressurised tanks behind a truck, which meant a 40 km/h limit, a safety light and slow-moving-vehicle signage. None of that was the fertiliser. All of it was the cost of switching to the fertiliser.

He did the same thing with automation. In 2002, the last year I worked on the farm, he put auto steer on the tractors. It was rudimentary by today's standards, expensive to implement, and it came with a learning curve. It also meant you stopped overlapping your passes, and when you are running a cultivator up and down 180 acres, a couple of feet of overlap every pass adds up to real money. The efficiency was there. There was still a cost to get to it.

That is the honest shape of what the agency owner above is signing up for. New skills, unbillable study, a different way of talking to clients, and a year where the model is half one thing and half another. Worth doing. Not free. Anybody who describes a repositioning as simply deciding to be something else has never paid for the hoses.

Putting it to work

Putting it to work

Run a loop, not a plan. Three moves, and the first one is the one you will skip.

Block the thinking time and defend it. Two hours a week, on the calendar, treated as client work, because it is. If your entire week is execution you have no mechanism for noticing that the market moved, and you will find out when a renewal does not renew.

Mine your sales calls for the AI signal. Go back through the last five discovery calls and write down every moment a prospect raised AI, in their words. Do you use it. I read something. Could we bring this in-house. Those sentences are your market research and they cost you nothing, because the calls already happened. Then decide what your honest answer is to each one, before the next call rather than during it.

Answer one question in writing: what does my market need from me this quarter? Not this year. This quarter. Write the offer that serves that need, price it, and put a date three months out in your calendar to ask the same question again. When the answer changes, change the offer. That is the whole discipline, and the calendar entry is what makes it a system instead of an intention.

One honesty check to run alongside it. List the things your clients pay you for that a platform feature could produce acceptably today. Not as well as you. Acceptably. That list is your transition plan, and the items on it are the ones you should be learning to teach rather than defending.

Sources

• Klaviyo, "AI Email Campaign Generator: Build and Send Smarter Campaigns", published 22 July 2026. Product page for Composer, in beta at time of writing; describes prompt-to-campaign generation including audience suggestion, and a standalone version for third-party platforms.
• Van Dycke, "How do you know if business advice applies to your business?" — on weighing a confident recommendation against your own architecture.
• Van Dycke, "If AI can produce the deliverable, what are clients paying you for?" — the companion argument on diagnosis as the scarce part of professional work.

Frequently asked

Will AI replace marketing agencies?

It will absorb the production steps that can be specified in a prompt, and campaign building is one of them. Agencies that sell execution by the hour are exposed. The work that survives is closer to teaching and judgment: knowing which move fits this business, and being able to bring a client's own team up to competence on tools that keep changing.

Should I switch from a retainer to a handover model?

Consider it, and give it longer than ninety days. Somebody at the client still has to own the work afterwards, and getting them genuinely capable takes closer to a year. Frame it as a transition from your team doing the service to you coaching theirs, and price the coaching honestly.

How do I know what my market needs right now?

Your prospects tell you on sales calls if you are listening for it. Write down every unprompted mention of AI across your last five discovery conversations. That is real, current, free market research, and it is more reliable than any twelve-month forecast you will read this year.

Isn't "stay adaptable" just advice to have no strategy?

A short loop is a strategy; it is simply not a destination. You commit fully to an offer for a quarter, sell it hard, and then genuinely re-ask whether the market still needs it. The discipline is in the re-asking and in being willing to drop something that is working less well than it did.

What if I get the next move wrong?

You will get some of them wrong, and a quarter is a survivable amount of wrong. That is the argument for the short loop over the long plan: a wrong three-month bet costs you three months, and a wrong three-year bet costs you the business.

Ready to look at the architecture honestly?

If you are trying to decide what your firm should be in a market that will not hold still, what helps is an honest read on what you are actually good at and what your market needs from you next. Nobody can hand you a forecast worth having. Twenty minutes, and we will tell you what we see. You can look at how we work through the four frameworks, or what an engagement involves, first.

Book the conversation →