AI for agencies - final blow?

Good news! Race to the bottom can be so much more efficient!!!

I wrote about agencies using AI a while ago. I talked about the necessity to change the pricing model and focusing on the outcomes, not ways we reach them. A lot has changed during that time, but the main thread remains. If you look at posts on LinkedIn, listen to smart people at big conferences - everyone is super excited about efficiency. They just fired their agency (copywriter, designer, etc) because look at what awesome content their new AI tool made. Savings, scale, speed!!!

I’m sorry, but it’s shit.

The reality is we’re using AI in overwhelmingly stupid ways. Trying to do more instead of doing better. Instead, I think the only (or at least straightest) way to get to peak amazingness is changing our mindset from “tool users” to “system builders”.

A great call from Y Combinator:

They see the need for companies that don’t just build AI tools for industry X, but instead design the full company in that industry around AI capabilities.

Applied to marketing, that means not building another tool that helps agencies write ads faster. It means rebuilding the agency itself around AI-enabled workflows. Not adding AI on top of existing structures, but redesigning the whole thing — from how insights are gathered, to how strategy is formed, to how creative is produced, tested and iterated.

Most of what we’re calling “AI adoption” today is just automation layered on top of broken processes. A faster way to do the same things. Full-stack thinking forces a harder question: if you were starting an agency today, with AI as a given, would you design it anything like the agencies we currently have?

Probably not.

While I was writing this, a new, spring 2026 “Requests for startups” dropped. YC doubled down on the subject, explicitly calling out agencies as ripe for full-stack disruption:

Meanwhile, in the marketing and advertising industry…

• We use AI to write ads

• We use AI to generate images

• We use AI to make endless variations and copies

• We use AI to summarise meetings, industry trend reports and whitepapers,

and underneath, everything stays the same. The peak of human technological development, used to make more of the same video, by an industry that’s slowly putting itself out of business. Trying to automate the simplest tasks (I’m fine with that, but…) and forgetting that there’s supposed to be another layer of thought before we start doing them.

My suggestion?

Redesigning the entire workflow around AI. Starting from a point of “what can we do better now”, instead of “where can we save money”. Focusing on how much more value we can create (and capture) instead of how many people we can lay off.

After tinkering with this idea for a few years, I admit that this is going to take some trial, error, and determination. But the most important part is the initial “prompt” for ourselves.

Here’s what makes sense to me right now:

Full-stack AI agency = connection of multiple AI services with clear human strategy on top, producing outcomes instead of assets.

I currently see it as 4 layers:

Layer 1: Inputs

That’s something that’s usually ignored. Mostly because it’s actually hard to do right. It’s not as glamorous and doesn’t make for a flashy case study. A lot of the time it’s not even about AI. Yeah, this part is “just” data: finding the right sources of it, cleaning it, making sure it’s uniform and not contaminated by noise. It can come from various places, but most probably you’ll be looking at

• Ads libraries,

• CRM,

• Performance metrics

• Customer reviews

• UGC

• Search trends

• Vertical-specific, Geo-specific, Platform-specific trends…

and many more, depending on your brand, audience and longterm goals.

Key here is to start small, and not try to have a perfect snapshot of anything and everything we think would be useful before moving to the next layer.

Layer 2: Intelligence

Patterns, hooks, insights, analysis… This is where AI can be tremendously useful. This is where, if Layer 1 is done right, we can start asking questions and “talk to the data”. That’s where strategy can start taking shape. Obviously, we need to know what questions to ask, as well as how to find the right insights and meaningful correlations. And it’s a great place for human strategists, who are now augmented by a much faster and insanely efficient partner that can surface the right, sometimes completely unexpected, things at the click of a button.

If your agency sells pretty pictures, and strategy is a free add-on to make pitch decks seem more intelligent, this won’t help too much. But if you see strategy as a value creating tool, then completely new frontiers suddenly open up.

Layer 3: Production

The stuff that’s most visible now. Usually done with very limited inputs (Layer 1) and, if we’re lucky, some random Intelligence (Layer 2). Often by separate people and without too much coordination.

Now, if done right this is where agencies could shine. Layer 1 and 2 should be the selling point. Layer 3 would be the amazing result of perfect execution earlier.

• Scripts

• Images

• Videos

• Audio

• AI-UGC

• Landing pages

• Email sequences

• Platform-specific and traditional campaign ideas

Not “look at how much we can do and how pretty it looks”, but “look at how well we know what exactly is needed and how effective it’s going to be”

Speaking about effectiveness - the final Layer 4: Improvement

is also frequently overlooked. That’s where performance should flow back into the system, seamlessly integrating with Layer 1 data. Models in Layer 2 take that into account, update their recommendations, correct earlier misconceptions and help the humans make sense of what’s important. Creative direction evolves and gets handed down, producing the perfect next iteration that Layer 3 then can bring into reality again.

Usually this last layer, as well as Inputs and, to some degree, Intelligence, were outsourced to Media or Digital agencies, trying to keep the connection with Creative only when necessary. Creatives were focused on the craft and “big ideas”, whatever that means. Fair enough - it’s hard to build out new capabilities and develop them to be competitive enough from scratch. That’s why it has clustered like that historically. I’d argue that, if used right, AI can help us to start meaningfully integrating again.

What this means for agency business models

Once you start thinking in systems instead of tools, a few uncomfortable truths show up very quickly.

First: the traditional agency model doesn’t fit anymore.

Retainers built around headcount make less and less sense when production is increasingly automated. Hourly pricing collapses the moment execution stops being the scarce resource. Selling “deliverables” becomes a race to zero when everyone has access to the same models.

Full-stack setups push agencies toward outcome-based models instead. You don’t charge for banners, videos or decks. You charge for impact. For growth. For accumulated knowledge and performance improvements that compound over time.

Second: team structures change.

You need fewer people doing manual production, and more people who understand systems. Fewer specialists working in silos. More hybrid profiles who can move between strategy, creative direction and data. Junior roles stop being execution-heavy and become insight-heavy much earlier in a career.

Creative directors shape feedback loops. Strategists sit inside live systems. Media stops being a separate black box and becomes part of the creative process.

Third: the competitive moat shifts.

It’s no longer about who has the biggest team or the nicest case studies. It’s about who learns fastest. Who integrates deepest. Who builds the tightest loops between insight, execution and performance.

That’s much harder to copy than a prompt library. And much harder to buy than a subscription to the same AI tool.

And finally: agencies that don’t make this shift slowly price themselves out of relevance.

If your value proposition is “we can produce more for less,” someone else will always produce even more for even less. The only defensible position is owning the system that turns data into decisions and decisions into results.

That’s the difference between being a vendor and being a partner.

So, to sum up… we come to a fork in the road.

On one side, we have AI as a feature, on the other - AI as an operating model.

AI as a feature is like swapping candy for nuts. AI as an operating model is a full diet plan. Both have advantages, but one has much more exciting long term possibilities.:

AI as feature:

• Faster decks,

• Cheaper copy

• More variation,

Basically, MORE of the same for LESS. Could be an attractive value proposition to the clients. Not sure for how long, and more importantly - how attractive it is for the agencies that want to be here in 5 years. This is easy to do - we don’t need integration and systems for this part. A basket of different tools is enough. That’s why everyone is doing it.

AI as operating model:

• Fewer people

• Tighter integration

• Continuous learning systems

• Outcome-based delivery

• and all the goodies of “AI as feature” sprinkled throughout.

This is much harder. But the upside is much higher, and the competitive “moats” are much deeper.

In my previous article I argued that in order to survive we need to make AI work for effectiveness, not merely efficiency. I think that this is how we can achieve that.

By designing systems that handle the boring, tedious and repetitive parts, we free humans to do what actually creates value: thinking, judgment and original ideas.

And this isn’t optional. Agencies that stay in “tool mode” will compete on speed and price until there’s nothing left to compete on. Agencies that build systems will compete on learning, outcomes and relevance.

That’s the fork in the road.