Creating and communicating a clear approach to AI adoption

This was not merely a question of enabling tool usage. It was a chance to create order from chaos for the wider team.

When generative tools started landing in designers' hands, the default org response was a vendor list and a Slack channel. That's not a strategy — it's an accident waiting to ship. Without a point of view, you get inconsistent quality, work that leaks where it shouldn't, and a slow erosion of the judgment that makes a design team worth having.

I wanted the opposite: a practice where AI accelerates the boring parts, makes the team's decisions more scalable, defines the risks when we have to proceed without certain inputs, and is explicit about who is accountable for what. The goal was never just "use more AI." It was "use AI with intention, and be able to own the output."

What we built

The practice took shape as a small, reusable kit that lived in a shared internal GitHub repo. I approached this as a product that the team could test drive and provide feedback on.

• Agents & apps for the repetitive, high-volume parts of the workflow — called to from the skills library during user interactions.
• skills.md workflows that encode how we (as Cisco IT Design) do things. This is information pulled from our organization's Confluence and vetted via language models to decide what information was already available to it versus what would be impactful if leveraged in service of our team.
• Prompt patterns shared and version-controlled, so good prompts compound instead of getting lost in DMs.
• Evaluation loops — the most important piece. A repeatable way to ask "is this output actually good, and good by whose standard?" before anything reaches a stakeholder.

The trade-offs I had to hold

How I knew it was working

• Adoption across silos. Designers outside my team picked the kit up and kept extending it.
• Faster, not sloppier. Concept-to-clickable-prototype time dropped substantially, while the evaluation loops kept the quality floor where it needed to be.
• It moved leadership's vocabulary. Leadership's conversations about AI in design changed shape after this, and we seemed to have more clarity about what short-, mid-, and long-term goals might be.

When we adopt new technology, we should understand how it might benefit existing process and change current ways of working. Without working through a period of experimentation and ambiguity together, individuals are left to their own devices.