If you run an agency right now, there's a good chance everything you're doing with AI is pointed at one idea, which is doing the same work you've always done but faster, or with fewer people on the team. That's the intuitive first move. A powerful new tool shows up, and you try to use it to take weight off the operation you already have. But the data coming out of PwC and McKinsey is making a pretty convincing case that this is actually the wrong play.

The companies getting real value out of AI are not treating it as a way to do the existing work cheaper. They're treating it as a growth technology. PwC found that the top fifth of AI-using companies are two to three times more likely than everyone else to say AI helped them find opportunities they wouldn't have found otherwise, and nearly three times as likely to say it helped them rethink their business model.

Efficiency is a race to the bottom

For agencies specifically, this matters a lot. If your AI strategy is just compressing timelines and cutting headcount, you end up in the same place as every other shop doing the exact same thing, and the only thing left to compete on is price. The agencies that actually pull ahead are going to be the ones using AI to do work they couldn't do at their size before, not the ones using it to grind out the same deliverables cheaper.

There's a reason individual productivity hasn't translated into organizational value. At most companies, everyone has their own AI habits, their own way of prompting, their own workflows, and none of that connects to anything else. It gets worse from there. A lot of people aren't even using the tools the way leadership asked them to, and at some companies employees are quietly sabotaging the rollout on purpose. Getting from individual productivity to something that actually shows up on the company's income statement requires building systems, which is a different thing than handing everyone a subscription.

"AI has made every individual 10x more productive, but no company has become 10x more valuable as a result."

George Savolka, a16z

What it looks like in practice

The best example I've seen of what this actually looks like is what Ramp (corporate cards & spend management) has built internally. Something like 99% of their employees use AI every day, which is a number you basically never see. The way they set it up, every new hire gets a fully configured workspace on their first day with all of the company's tools and context already wired in, so the AI has something to work with from day one instead of starting from zero.

When someone on the team figures out a better workflow, that skill gets published back to the whole company. They have hundreds of shared skills now, and an internal marketplace that surfaces the right ones to the right people based on their role.

What's interesting about their approach is that they specifically decided not to simplify AI for non-technical users, which goes against most of the default advice right now. Their argument is that when you pair the full capability of the tools with AI's ability to help people work through problems in real time, most people can figure out how to do serious work with them. You don't have to dumb the tools down, because the AI itself becomes the teacher.

What this means for agencies

You're not going to build what Ramp has built, and you don't need to. The principles scale down. Your job as an agency owner is to help the people on your team actually connect their tools together so the AI they're using has context about your clients and your projects, instead of making everyone start from a blank slate every time they open it. And when somebody on the team figures out a workflow that works well, that should become a company asset, not a piece of personal tradecraft living in one person's head.

The real shift is the one between using AI to do your existing work faster versus using it to do work that wasn't possible at your size before. Agencies that treat AI as an internal operating system, something that compounds as the team learns from each other, are going to end up in a completely different place than the ones still using it to shave a couple hours off the same timeline they were running last year.