A few weeks ago I wrote that AI has moved upstream — that the most capable AI systems no longer wait to be asked. They initiate. They act. The question that post left open is this: what happens when your agency is still organized around work that starts downstream?
My friend Aby Varma puts it simply: "AI can help you do things differently. But it can also help you do different things."
Those two sentences describe entirely different futures. And right now, most agencies are choosing between them without realizing a choice is being made.
This week, my colleague Teri Sun and I ran a workshop at the Healthcare Marketing & Physician Strategies Summit (HMPS) built around a gap I keep seeing: the distance between AI experimentation and AI systemization. Teams that experiment are everywhere. Teams that have actually scaled it — turned individual experiments into connected, repeatable workflows — are rare.
Experimentation is the first path. Using AI to do familiar work faster, clean up copy, summarize meetings, generate a dozen variants where you used to write one. Real value. But it's still the same work, done differently. You're downstream — moving faster in the same current.
The problem with stopping here is subtle. Add enough point solutions to a fragmented workflow and you don't get transformation. You get sophisticated chaos. Wrike's research published last month found that 96% of knowledge workers say it would be valuable if their AI tools shared context and worked together. They can't. Not in most agencies.
McKinsey's April report on agentic AI named the result: "a patchwork of disconnected pilots and systems that increase activity while delivering few meaningful enterprise-wide benefits." The gen AI paradox — AI is everywhere except on the bottom line.
This is the death by a thousand cuts I'll be discussing at the Build a Better Agency Summit on May 19. Disconnected adoption happens one tool at a time, and before you know it, your team is prompting from scratch on every project, duplicating effort nobody can quite see because it's buried across a dozen platforms.
Doing different things means going upstream. It means asking what your agency is actually for when the execution is handled — when production, templating, summarizing, and first drafts become things AI does faster than any human could.
What do your people do with that time? What do you sell that a client's own AI stack can't replicate? What does your value proposition look like when "we produce great work quickly" becomes the price of admission rather than the pitch?
The agencies pulling ahead aren't using more tools. They're using connected ones — built on a shared foundation of knowledge and workflow that compounds over time. Institutional memory that doesn't leave when an account manager does. Humans freed to do what actually requires a human: judgment, relationships, strategy that knows why, not just what.
That's AI as infrastructure. Not a layer on top of the current, but a different river entirely.
The two paths are a sequence, not a choice. You almost always have to do things differently before you can do different things. The problem is treating the first as the destination.
Less than 10% of marketing organizations have captured value across end-to-end AI workflows. Ten percent. The ceiling is wide open.
The agencies that move through both phases — freeing up capacity first, then fundamentally rethinking what to do with it — are the ones that will look back on this moment as an inflection point.
The rest will realize they had all the tools. They just never went upstream.
Matt Cyr is the founder of Loop, an AI transformation consultancy for marketing agencies. He'll be presenting "Death by a Thousand Cuts: The Hidden Cost of Disconnected AI" at the Build a Better Agency Summit on May 19.