Putting AI to work inside a mid-sized agency. Without the panic.

Two years ago, the AI conversation inside agencies was mostly defending existing workflows or doomscrolling. My approach was less interesting and more useful: figure out what AI could actually do for the work, do it, and spread what worked across the company.

At 1021 Creative, I founded the AI Task Force, a cross-functional group with a mandate to ship AI pilots into actual workflows. We started with a company-wide survey to understand where people actually were on AI: how aware, how comfortable, and how much they were already using on their own. We wanted to know what was happening in the building, not what was being said about AI in strategy meetings.

Then we worked through the company’s functions in priority order, with pilots owned by the function leads who’d actually have to deploy them. The first pilot was Editorial AI Gems, a content-production tool we shipped into the editorial workflow. From there we rolled AI capabilities into research workstreams, content enrichment, and the proposal/pitch process. The pilots that survived were the ones that held up against real client demands, not just committee pet projects (gotta kill your darlings sometimes).

The biggest single build was TIIP, the agency’s proprietary trend intelligence pipeline. It started as a manual Airtable submission-based process that didn’t scale. The next-generation system we architected ran on Postgres with pgvector for embeddings and used clustering to detect emergent patterns in cultural and behavioral signals. TIIP became the engine underneath the company’s largest client workstream and produced meaningful efficiency gains in throughput and turnaround.

The “without the panic” part isn’t just rhetorical. Change management is hard. We designed the work to land as upskilling rather than displacement. The Task Force was cross-functional, not a centralized AI team brought in to replace people’s work. Pilots were owned by function leads. And the company’s upskilling program ran with HR as a voluntary center of excellence, not mandatory training. The result was an organization that adopted AI capabilities at its own pace, with the people most affected by the work running it.

The strategy underneath it all was build vs buy: use commercial tools where they’re stronger, build proprietary capabilities where the commercial tools fall short. I secured initial budget and structured the work in phases, with later phases funded based on how earlier phases performed.

None of this was AI for its own sake. The work made a 250-person global agency faster and better at the work it already did, while building capabilities it didn’t yet have.