AI

I spend my days as the Head of AI and Development at a marketing agency, a role that affords me the opportunity to navigate an endless array of corporate “flavors”. You don’t want to know how a company’s flavor is determined. Taste tests are a hazard of the job.

Lately, I have been trapped in a cognitive mosh pit. On any given morning, I am celebrating with my team as we deploy agents that people call “their new BFF” and I feel like I am rewriting the laws of physics. But by the afternoon, I am reading a report about the latest amazing AI breakthrough and I feel like I’m playing with Duplo blocks at the kids’ table.

It is a constant, exhausting oscillation between delusions of grandeur and deep-seated imposter syndrome. On the plus side, you do get cupcakes at the kids’ table.

The clarity came during a conversation with a friend. He noticed that the “AI Revolution” we read about is actually two different stories. One story is about the “Garage” – tiny teams of fewer than 5 people moving at the speed of light. The other is about the “Empires” – companies with $5 billion in liquidity who can simply buy their way into any capability.

Neither of those stories applies to the “Messy Middle.”

The middle ground is where most of the world lives: companies with more than 5 people, less than $5 billion, and more than 5 products to keep alive. In this space, you don’t have the luxury of pure speed, you don’t have the luxury of pure scale, and you don’t have the luxury of pure customization. You have to find another way.

If the muse stays with me I can write many more articles about this topic, but here’s at least an opening salvo.

TL;DR

Most AI advice is written for 5-person startups or billion-dollar giants. This is for everyone in between (the “Messy Middle”) where speed, scale, and budget all have limits. After deploying AI across a marketing agency, here’s what actually moves the needle:

• Stop mistaking busy-looking AI for real productivity.
• Find the specific leverage point for each team.
• Use AI to reduce the coordination tax that quietly drains organizations.

Lesson 1: Adoption is Productivity Theater

We achieved 100% adoption of AI coding tools a long time ago, and for a moment, I thought I had hit the jackpot. It wasn’t quite the billion-dollar Powerball, but it still gave me the warm fuzzy feeling of a winning ticket. Every developer had a magical army of faithful AI assistants at their command, like rooks on a chessboard. They were all telling me the same thing: “I’m saving 50% of my time!”

And yet, when I looked at our project reports the actual hours spent on our development projects remained stubbornly, depressingly, the same. To quote my own reaction at the time: “Huh?!?!?!”

The answer was hiding in plain sight. “My AI is thinking” had become the new “my code is compiling.”

Back in the stone ages, we had to wait for our code to compile before we could test a change. At one company I worked at, a full rebuild took over an hour. That was an hour where the computer was unusable, which is one reason why tech companies started putting ping pong tables in their break rooms. With AI, developers were indeed saving “hands-on-keyboard” time, but they were immediately spending that time watching a progress bar or waiting for the model to finish its thought. It was productivity theater. It was a mirage.

The business was gaining exactly as much value from those five minutes of “AI thinking” as my old company gained from my mid-compile ping pong matches. Which I won upwards of 40% of the time, by the way.

While this logic is easily applied to developers, it is a trap that catches every department in the agency. If we save a project manager five minutes on their note-taking, that is a win for the PM but it is not a win for the business. You cannot simply squeeze more business-relevant work into a five-minute window. Without a fundamental change in the nature of the work, they might as well be playing ping pong.

The hard truth: If you are using AI to do the same tasks slightly faster, you aren’t transforming; you are just subsidizing downtime with productivity theater.

Lesson 2: Find the Leverage Point

Archimedes said, “Give me a lever long enough and a fulcrum on which to place it, and I shall move the world.” To move beyond productivity theater, you have to find the leverage point (the fulcrum) for your team and your workflows. If you put your primary effort in the wrong place, nothing will budge.

As I have worked with a plethora of teams at my company and others, I have noticed a few commonalities for finding the leverage point.

1. Major time savings: “This would save me hours, not minutes…”
2. Improved quality: “I wish we had time to do this better…”
3. Increased output: “If this didn’t take so much work, we would do it way more often…”

Major time savings. When we first offered AI and automation help to the whole agency, we immediately received hundreds of requests. To help us prioritize, we set a time savings benchmark of 20 hours per month, and ideally focused on one or two people. Anything smaller than this should be prototyped by the requesting team instead of getting the AI engineering treatment.

But why the emphasis on one or two people? If two people save 10 hours per month each, they might be able to take on a few extra tasks or support an additional client or get to important tasks that have been languishing. If those same 20 hours are spread among 40 people, we’re back to playing ping pong in the break room.

Improved quality. Most people want to be good at what they do, but excellence is often buried under a mountain of low-value grinding drudgery. What if we could let AI take more of the drudgery and let the humans unleash their creativity on the things that differentiate us? Here are some real examples:

• For Digital Designers: We aren’t typically using AI to “make art.” We use it to clean up messy Figma files, find hard-coded values that should be design tokens, and bridge the gap between design and code. We automate the documentation so the designer can stay in the creative flow.
• For Data Analysts: We move away from manual slide-building and toward conversational business intelligence. We build reporting pipelines that allow them to spend their time extracting insights instead of wrestling with Excel formulas.
• For Account Managers: We build tools that consolidate information across sprawling accounts with dozens of projects. Instead of chasing people for updates, they have instant, broad visibility.

The key point here is to find the right leverage point. We had to build tools for a different stage of the workflow for each example above. If you try to have one single fulcrum for every type of world you’re trying to move, you will have some shiny successes and wonder why these other teams aren’t budging.

Increased output. Agencies often spend a lot of time defining the limits of our work – we will do two visual directions and flow the content into a website exactly once for up to 35 pages. As AI tools improve, we are beginning to think more about abundance – how much more can we offer our clients – rather than limiting our exposure.

Our SEO team has an amazing, in-depth analysis they do to ground their recommendations in reality. When we have this analysis, we have clear visibility and guidance. In the spirit of increased output, building AI tools around this process allows us to offer it to more clients, to increase the number of competitors we consider, and to re-run the analysis periodically instead of limiting it to the start of our engagement.

On another project, we’re working with a startup that is still building the bridge while they’re crossing it. In the past, I would have placed so many limits on them to avoid a thousand re-writes. But now that prototypes built by AI are so fast, we are able to build dozens of iterations to help them find their product and process.

If you can find a spot to offer more, to sell more, to analyze more, or to loosen your restrictions, that’s a good AI leverage point. And if you want your CFO to materialize out of thin air, just say the words “recurring revenue” aloud three times while facing the rising sun.

Lesson 3: Cut the Coordination Tax

I’m going to say something that might get me barred from certain corporate retreats: “Meetings are often the most expensive way to solve a problem.” Ducks in anticipation of sharp objects being thrown at me… Oh, you all agree with that? Cool.

For those of us in the Messy Middle, a lot of time and energy is spent coordinating different teams. Processes that work for a company with three people don’t work for a company with 300 or 3000. But what if we could reduce the time, energy, and cost of that coordination?

We have achieved this with two broad strategies. The first was already mentioned above: making information retrieval easy. If someone needs to commission an Indiana Jones-style archaeology expedition to dig up information, you’re going to wind up with a lot of meetings. When we rolled out a fully connected system across all our platforms, people could just get reliable answers with an AI chat instead of scheduling a “quick” check-in meeting. Multiply the non-existence of that meeting by however many people would have been on the invite, and you get business-relevant savings in a hurry.

The goal is to make information so accessible that “checking in” feels like a waste of time.

The second strategy to make collaboration less expensive is to make tools that allow one person to cover more territory. A back-end developer with heavily-vetted front-end coding tools doesn’t need to hand the work over to someone else, so that’s another meeting that doesn’t happen. A designer with a specially trained brand-aware AI writing tool can add draft content directly to the design instead of placeholder content. That’s one less dependency in the project plan.

When a single person can expand their reach into the entire lifecycle of a task, the coordination tax vanishes. You don’t need a handoff meeting if there is no one to hand it off to. We aren’t shrinking the talent – we’re shrinking the drudgery.

Every time a pure coordination meeting doesn’t happen because of my AI tools, I smile approvingly like a vanquishing general.

Summary

AI in the Messy Middle isn’t about chasing the frontier — it’s about finding where effort actually compounds. Adoption alone is productivity theater. The real wins come from identifying high-leverage moments for specific people, using AI to elevate the quality of human work rather than just accelerate the same tasks, and systematically eliminating the coordination overhead that organizations quietly accept as the cost of doing business.