What AI Actually Changed For Marketing Agencies

Our co-founders, Razy Shah and Dhawal Shah, have run 2Stallions together since 2012. In that time they have been inside the marketing of hundreds of companies, from startups to MNCs.

Razy sat Dhawal down for the first episode of the relaunched Marketing Expert Series and asked what AI has genuinely changed in marketing, and what is still hype.

Dhawal’s answer was not the one Razy expected, and it was uncomfortable in places. The short version: the speed of execution has collapsed, judgement has not moved at all, and an agency built on the first of those is in trouble.

What genuinely changed: speed and efficiency

Dhawal’s example is one the team at 2Stallions remembers well, having watched it happen.

Five or six years ago, the agency ran an in-depth SEO analysis covering thousands of pages. Dhawal spent three to four weeks with the team going through them, writing scripts in Excel to find the patterns, then working out what the insights actually were.

The same piece of work now takes a couple of days. The output is better, not worse, because the time that used to go into wrestling data into shape now goes into interpreting it.

That is the real change, and it is not confined to marketing. Drafting, reporting, analysis, all of it moved.

What did not change: judgement

Here is where the conversation turned, and where it stayed for the rest of the episode.

The loudest claim in marketing right now is that autonomous agents will do all of it. Dhawal is on that train himself. He is actively experimenting, he would happily hand the work over, and he still does not think it is possible yet.

“AI is a tool that you as a user wield, and if you don’t have the right judgement, your output is still gonna be pretty poor, regardless of how good the AI is.”

Garbage in, garbage out. The phrase is old and it has not stopped being true.

Why this breaks hourly billing

This is the part that should worry anyone running an agency, and it is the part 2Stallions has not fully solved either.

The standard agency proposal prices work by hours. Ten people, this many days, therefore this much money. That was a reasonable proxy when hours and outcomes moved together.

They no longer do.

If a piece of work took three weeks and now takes two days, and the client outcome is the same or better, then the hours were never the thing being bought. They were a stand-in for something else, and that something else is judgement.

“You can’t bill by hours anymore… now with AI your speed of execution has increased dramatically, so the way you have to price it is based on your judgement.”

What clients are actually paying for:

• Knowing what the spec should be before anyone starts
• Knowing when the output is wrong even though it looks finished
• Knowing which of the forty things AI just produced is the one worth sending

Value-based pricing has been discussed in this industry for twenty years and mostly stayed a discussion. What changed is that the hours model now actively misprices the work rather than merely approximating it badly.

2Stallions has not landed on the answer yet. Neither, as far as the team can tell, has anyone else. But the clock is running.

What happens to agencies that only sell execution

Razy asked Dhawal directly: how does a digital marketing agency, in its current shape and form, survive the age of AI?

"If you stick to the current shape and form, then no, you don't survive."

Dhawal's reasoning is straightforward. A lot of agencies today are focused purely on execution. 2Stallions was in that position itself about a year ago, so this is not a comment aimed at anyone else. If execution speed is the thing you sell, you are in a contest you cannot win, because your speed can never match what AI can do.

The shift is in what the agency sells. Strategy and judgement, rather than hours of output. And the hard part is not the strategy, it is people accepting that their role has changed.

"It's a mindset shift in terms of what your role is, and you need to be comfortable with the new role. That's where the crux of the pain is."

Why your AI content sounds generic

A more immediate problem, and one every marketing team has hit.

You ask AI to write a post. It comes back readable, on-topic and completely characterless. The instinct is to blame the tool.

Dhawal's view is that the expectation is wrong.

"They just assume, hey, I'm gonna give it a one-line prompt and it's gonna give me something magically… Your AI still needs to be trained, and the AI still needs good quality examples."

The usual failure is that the brand voice lives on a slide deck nobody on the team has opened in a year. It is not in the prompt, so it is not in the output.

What to do about it this week:

1. Find your brand voice guide, wherever it lives
2. Put it into the model as context
3. Add three to five examples of writing that is right, and say why each one is right
4. Add two or three examples of writing that is wrong, and say why
5. Ask the model to produce a voice guide from that input, then save it and reuse it

That alone changes the output more than any amount of prompt tinkering.

Working out where AI belongs in your marketing?

That is a good deal of what 2Stallions does now. Talk to us about your digital marketing and we will tell you honestly which parts of it we would automate and which parts we would not.

Where to start with AI agents

Everyone is trying to build agents. Dhawal's advice is to start somewhere else entirely.

"Do not start by building agents. Start by finding a problem that you want to solve."

He tried it the other way round first. He wanted an agent that would run his entire LinkedIn presence while he talked to it, and nothing more. It fell over quickly, and the useful part was understanding why.

So he went back and did the slow version:

1. Worked out what the problem actually was
2. Built the process by hand in a Claude project
3. Found the two or three points where his own judgement was genuinely required, rather than merely habitual
4. Only then moved pieces of it into something more autonomous, in Claude Code

The result takes him twenty to thirty minutes a week and still sounds like him, because he knows exactly where he needs to be in the loop and where he does not.

Every marketer 2Stallions knows getting real value out of AI right now did the slow version first. Every person still waiting for the tool to solve it skipped that step.

What to ask before hiring a marketing agency in 2026

The old evaluation criteria have stopped working. Proposal quality, idea quality, turnaround speed, every agency now produces a good proposal quickly, because every agency has the same tools.

Two things separate them now.

Strategy. How good is it, how deep does it go. AI cannot do this part well, so a human has to.

How they actually use AI. Not whether they use it. What for.

"If you ask an agency what AI tools they use and what work they do with AI, and all they can say is 'we use ChatGPT,' that's probably a red flag."

You want it used for processes, analysis and drafting, not only for generating content and visuals.

And ask about data. If an agency is running your brand guidelines, internal memos and campaign ideas through AI tools, you want to know those tools are not training on your data and that someone has thought about leakage.