There Is No Such Thing as “AI Adoption”

When a new technology arrives that touches everything, the worst thing you can do is treat it like one thing.

The Web Was One Channel. AI Is Not.

When the web arrived, it was disruptive, but it was also bounded. You had to figure out what your website said, how it was built, and whether people could find it. The strategic challenge was real, but the container was clear. There was a place where the web lived. You went there, you sorted it out, and the rest of your agency kept operating more or less the same way it always had.

AI does not work like that. It reaches into writing, research, project management, estimating, client service, business development, creative production, and now software development. It doesn’t have a home address. There’s no “AI section” of the agency to renovate while everything else stays put. Everything you build from here on gets built with different stuff, and the question is whether you’re making decisions about it or just letting it happen.

So when someone asks “how do we adopt AI?” – which is what most owners are asking right now – the question is too coarse to answer well. The better questions are which parts of your agency have the most friction, where your margins are thinnest, which capabilities you’re missing, and what you can realistically change in the next ninety days without breaking the client work that pays the bills.

The Problem With a Broad Initiative

Most agency owners who try to tackle AI as a single initiative end up with scattered experiments that don’t compound. One person on your team has figured out a great prompting approach for first drafts. Someone else built a research workflow they love. A third person is using AI for social posts. None of it connects, none of it gets documented, and the learning stays at the individual level rather than becoming organizational capability. Six months later, you have a pile of individual habits and no actual operating improvement to show for it.

The other failure mode is the opposite: you bring in someone to “do AI” across the whole agency at once and end up with a big training day, a folder full of prompts nobody uses, and a lingering sense that you spent money without changing anything. Training is not the same as adoption. Adoption means the workflow changed, the handoff changed, the output standard changed, and your people know what they’re responsible for.

Neither scattered experiments nor a single sweeping initiative solves the real problem, which is that AI adoption is a change-management challenge dressed like a robot. The technology is the easy part. The hard part is figuring out which capabilities are most important to your specific agency, establishing the right sequence for building them, and then doing the boring work of embedding them into real workflows with real review standards.

Five Areas, Not One Initiative

Breaking AI work into five areas rather than one broad initiative forces a useful decision at the start: which of these is most relevant to your agency right now, and in what order? The five areas are operations, marketing, creative, tool-building, and an honest diagnostic of where you currently stand. An agency can concentrate on one or move through several. The point is to start where the pain is, not where the technology is most exciting.

This is the structure behind how Ironwood AI works with agencies. Each area maps to a focused lab, because treating everything as one engagement is how you end up with a big deliverable that doesn’t change anything. Agencies come to us for one lab or several, depending on where the friction is. But if an agency hasn’t gotten started at all – no real audit, no clear picture of what’s already being used or where the gaps are – we usually begin with the Opportunity Lab, which is a diagnostic phase: interviews, workflow assessment, and an honest read on which of the other areas are worth tackling first and in what order. Without that, the priority list gets jumbled.

Operations goes behind the scenes: sales workflows, estimating, project management, client service, knowledge systems, delivery processes. This is where a lot of the quiet margin lives. If your estimating process relies on gut feel and a spreadsheet that someone updates by hand, that’s a fixable problem. If your project kickoffs are inconsistent because everything lives in someone’s head, AI can help you build systems that are repeatable whether or not that person is in the room.

Marketing focuses on your own pipeline. List building, outreach, LinkedIn, email, content, business development – the things agencies are great at doing for clients and bad at doing for themselves. AI removes several of the real friction points without turning your outreach into the kind of robotic mass messaging that makes everyone feel worse about the internet. The goal is a business development system that genuinely runs, not a plan you revisit every quarter and then abandon.

Creative is where agencies often want to start but often shouldn’t – at least not before operations and marketing are sorted. The creative work is where AI is most visible, but it’s also where undisciplined use does the most damage to your credibility. The focus here is on getting dependable results, maintaining voice and brand standards, and understanding the difference between work where AI adds real value and work that produces output your editors have to fight. Writing, imagery, video – each of these has a different risk profile and a different set of practical techniques worth building into your process.

Tool-building is the one that surprises people. Agentic AI tools like Codex and Claude Code have made it realistic for non-developers to build custom lightweight applications for their own operations – a brief intake tool, a QA checklist, a proposal generator, a client-facing widget – without waiting for a SaaS vendor to ship the exact thing you need. This is not for every agency right now, but for shops that have real workflow problems that no off-the-shelf tool solves well, building something specific to your needs is no longer the long, expensive project it used to be.

What You’re Actually Deciding

A ten-person shop with a thin pipeline and a business development process that amounts to “wait for referrals and occasionally post something on LinkedIn” has a different first priority than a fifteen-person shop with solid new business but a creative process that’s inconsistent and hard to scale. The first shop probably starts with marketing. The second probably starts with creative or operations. Both of them should start with a plain audit of where AI is already being used, where policy is missing, and where the biggest workflow friction sits – because without it, the priority order is a guess.

The point isn’t to work through all five areas in sequence or to treat AI adoption as a project with a start and an end. The capability compounds if you build it deliberately. It evaporates if you let it stay fragmented.

Pick the area where the cost of not changing is highest. Start there. Get it working before you move on.