> ## Content Index
> Fetch the complete content index at: https://vill.ar/llms.txt
> Use this file to discover other available public pages before exploring further.

# How the modern agency will evolve
- URL: https://vill.ar/how-the-modern-agency-will-evolve/
- Published: 2026-11-05T17:35:24.000Z
- Updated: 2026-11-06T04:05:05.000Z
- Author: Mike Villar

## The short version

![Illustration of falling papers on execution cost scale, rising larger paper on judgment value scale.](https://vill.ar/content/images/2026/11/poe-24eaba90.jpg)

Execution is getting cheap, and judgment is getting expensive. Agencies that sell hours of production will get squeezed from both sides, by clients who can generate a first draft themselves and by competitors who deliver ten times the output at the same cost. Agencies that sell decisions, taste and accountability will do fine. Everything below is the long version of that sentence.

### The client math has changed: expectations expand, patience contracts

Clients have seen what a chatbot can do in thirty seconds. So when we tell them a content calendar will take two weeks, they hear a number that doesn't match their experience. Their expectations of scope, volume and turnaround keep growing, and their patience for the old pace keeps shrinking.

You can complain about this or you can price it in. I've stopped complaining. The client who asks why a landing page variant takes three days isn't being unreasonable. They've simply recalibrated, and we're the ones who have to catch up.

The money is already moving. The Stensul 2026 MarTech Outlook found that organizations are treating AI as a core investment, with budgets rising and teams shifting work in-house. That's the buyer's side of the same story. They're building the muscle to do more themselves, and every task they pull in-house is one you can no longer bill for.

### The new bar: be better than what clients can do themselves with AI

Here's the test I use now. If a smart marketing manager with a good model subscription and an afternoon could produce your deliverable, you don't have a service. You have a demo.

I wrote about this in a piece on [narrow-scoped software](https://vill.ar/the-death-knell-for-narrow-scoped-apps/), and the same logic applies to services. If a client can strip away your logo and find the agency equivalent of a form and an empty database, they'll replace you the way I've been replacing plugins and subscriptions with my own vibe-coded products. What survives is the stuff an agent can't reverse engineer in a few minutes: judgment, taste, accountability, and a track record of being right when it counted.

The bar is no longer "better than the agency down the street." It's "better than what this client gets by doing it themselves with the same tools." That's a much higher bar, and a lot of agencies haven't noticed it moved.

## What AI can't take: accountability, agency and skin in the game

I'm bullish on AI and I'll say so loudly, but four things don't transfer to a model: accountability, agency, decision-making and skin in the game. They won't disappear. They'll take different forms, and your job is to make sure you're the one holding them.

A client pays an agency for a named person who stands behind the work. When a campaign tanks, a model doesn't take the call, absorb the awkward silence or lose the retainer. A human does. That exposure is the product, and it's why clients don't fear AI-assisted content. They fear content that nobody stands behind.

I built this into [Poe](https://vill.ar/poe-the-ai-hub-for-the-modern-agency-content-team/), the AI platform we use for content at Growth Rocket. A person approves every idea before it becomes work. Every guideline suggestion is applied, adjusted or dismissed by a human. That's a design decision, not a gap I haven't fixed yet.

### Which agency work keeps its value

The cleanest way I can split agency work is judgment versus throughput. Strategy, creative concepting, negotiation and accountability sit on the judgment side. Campaign builds, resizing, versioning and routine reporting sit on the throughput side, and their market price is falling fast.

Industry analysts see the same split. ALM Corp's breakdown of AI's squeeze on agencies says AI isn't replacing agencies outright, but it is devaluing the parts of agency work that used to support the pricing. That matches what I see in client conversations. Nobody argues about paying for a good strategic call. Everybody argues about paying for the fortieth ad size.

The holding companies show where this goes at scale. Gradient Group's analysis of the big networks argues that the most important word in the story isn't layoffs, it's leverage. They're building businesses that need fewer people per dollar of revenue. Omnicom says AI will affect both staffing and pricing, WPP is automating high-volume creative and production through WPP Open, and Publicis reports nearly its whole workforce working with Copilot.

### Marketing is part art: the artist-with-a-pencil argument

Give a pencil to a non-artist and you get a stick figure. Give the same pencil to an artist and you get a portrait. AI is the pencil, and the output depends on whose hand is holding it.

Digital marketing has always been sold as a science, with dashboards, attribution models and A/B tests. Those matter, but real digital marketing is part art. Knowing which angle will land with a skeptical CFO, when a headline is too clever, or which of five technically correct options feels right: that's taste, and taste comes from years of seeing what works and what doesn't.

Almost everything written about AI and marketing treats it as a pure efficiency story. I think that misses the main point. When everyone has the same pencil, taste is the only thing left to compete on.

## The internal reckoning: who survives the great flood

Now the uncomfortable part. Inside agencies, there will be a major unprecedented blood-letting, and I think it's inevitable. I'd rather say it plainly than pretend we can all hold hands and upskill together.

The numbers at the top of the industry are already ugly. WPP lost roughly 9,400 staff in a single year. Omnicom absorbed IPG and announced 4,000 redundancies. Forbes reports that AI agents are displacing junior marketing roles at major firms, citing Forrester forecasts of further cuts. Smaller agencies like ours won't make headlines when they do the same, but the math applies to them too.

## Adaptation is not optional

### I'd take AI-native junior engineers over tenured ones who won't adapt

I'll say the controversial part out loud. I would rather have a fresh junior engineer who can orchestrate AI well than a tenured engineer who can't work with it.

I'm not a developer, and I've shipped real things anyway. [I vibe coded the Growth Rocket website redesign](https://vill.ar/trusting-vibe-coding-with-a-major-redesign-project/) from a plan I painstakingly wrote and revised an embarassing number of times. I built Poe the same way, directing a mix of AI agents and keeping the decisions, and all the "no, do it again" prompts for myself. Both projects spanned a couple of days. Projects like these once meant handoffs between design, front-end, back-end, QA and SEO, and a timeline measured in quarters.

If I can do that without writing code, imagine what a sharp young engineer can do. Years of tenure matter less when the tool compresses the learning curve. What matters is whether you can break a problem down, direct the agents and spot when the output is wrong.

### I'd take AI-native writers over purists who hand-edit everything

Same logic for writers. I'd rather have an AI-native writer who can produce high volume at high quality than a Pulitzer-level writer who insists on spending most of the day hand-editing.

That sounds like an insult to craft, and I don't mean it as one. Craft still matters. But if the craft is deployed on the wrong layer, as sentence-by-sentence polishing of text a model could have drafted, it's a luxury a client won't pay for. The AI-native writer puts craft into the brief, the structure, the rules and the final read.

We learned this building Poe. The same written rules should instruct both the writer and the reviewer, because AI is only as consistent as the rules people give it. The writers who thrive are the ones who can write those rules.

## The honest counterpoint: what we lose if we stop hiring juniors

I'm aware of the problem with my own position. If every agency swaps junior hires for agents, where do tomorrow's senior people come from? The Content Marketing Institute makes this point well, warning that cutting entry-level hiring risks long-term trouble. If you replace the junior with a prompt, you eventually run out of people who learned the craft by doing it.

We've written about the ethics at Growth Rocket too, because replacing junior marketers with AI agents creates real dilemmas around career development. I don't have a clean answer. My view is that we should hire juniors differently, not stop hiring them. Bring in people who are AI-native from day one and train them on judgment, client handling and taste, since those are the skills no model teaches them.

That means the entry-level job changes shape. It's less production, more orchestration and review. It's also a bit harder, because the junior has to be right about things earlier in their career.

### Why product managers have an edge when code is no longer the barrier

I think product managers are in a strong position. Their whole job has been turning fuzzy needs into precise specs, and now that technology and coding no longer stand in the way, that skill pays off directly. Reforge makes a related argument: AI automates or eliminates the low-value parts of product management and amplifies the pieces that always separated great from average.

I've felt this firsthand. The projects I've directed with AI agents worked because I wrote a plan, reviewed it, and held the line on decisions. I wasn't coding. I was specifying, and then saying "no, do it again" until it was right.

If you're building an agency team, look for people who think like product managers, whatever their title. They ask what problem a thing solves, they write down what done looks like, and they notice when the output drifts from the intent.

### Knowing how the sausage is made beats knowing how to make it

You don't need to be able to make the sausage anymore. It's enough to know how the sausage is made.

![Illustration of sausage-making machine cross-section with hands inspecting finished sausages.](https://vill.ar/content/images/2026/11/poe-08d4850e.jpg)

That means you understand what good code looks like, basic DevOps, how parts of a system talk to each other, how a new feature fits into the puzzle, what a solid SEO audit covers, and why a campaign structure works, even if you can't build any of it from scratch. That knowledge lets you spot the bad sausage when the machine hands it to you. A person who has never seen the process can't tell good output from a confident-sounding mess.

This is why I don't think deep expertise is going away. It's changing jobs, from the person doing the work to the person who knows when the work is wrong.

## Either command the model or compete with it

Here's the line from my own notes that I keep coming back to:

> There is no universe where a human can compete with a model that has billions of parameters and detailed instructions, unless that human is in absolute and total command of that model.

A model reads faster, drafts faster, and never gets tired at 4 p.m. on a Friday. Racing it on production is a losing bet, and most of the resistance I see inside agencies amounts to people entering that race.

The winning move is to command. Know the model's strengths, give it clear instructions, set up the checkpoints and own the result. Treat AI as a team of specialists, each with a role, each reporting to a person who answers for the output.

That isn't a diminished role for people. I'd argue it makes people more in the loop than before, with more responsibility per head and fewer places to hide.

## What this means for how agencies hire, price and organize

On hiring, I screen for judgment, taste and AI fluency first, and credentials second. I want to see how someone works with a model before I look at their résumé. Tenure counts only if it comes with a willingness to change how you work.

On pricing, hours are a dying unit. If a task takes a model four minutes, billing for four hours is a conversation you'll lose. Price on outcomes, decisions and accountability, and be honest with clients about what the AI did and who approved it. Clients respond well to that transparency.

On organization, expect smaller teams with more senior leverage. Your best editors become operators directing a team of AI specialists, accountable for every word that ships. Fewer layers, shorter handoffs, and a clear named owner for every deliverable.

Some agencies will resist all of this, and some will get acquired or shut down. The ones that adapt will be smaller, sharper and more accountable than the agencies we grew up in. I think clients will be better off for it, and I plan to be one of those agencies.