Case Study: AI
How I'm helping teams at Gusto ship with AI tools, from new coordination primitives to frameworks that make visions buildable.
Context
I lead design for Core Experiences, the platform team that makes Gusto feel like one connected experience instead of a bunch of separate tools.
I'm head of design for Core Experiences at Gusto. I have 5 designers and sit in the group's platform leadership alongside a Sr. Director of Product and Sr. Director of Engineering.
I started my career as an engineer, so the platform and AI work I'm about to walk you through is a fun, full-circle moment.
Making visions shippable
Some teams were creating incredible visions but couldn't start building them. Others were shipping constantly but the work wasn't adding up to anything bigger.
Right now there's real pressure for a step change in product delivery. With AI tools, leadership is asking "why not now?" instead of planning on a traditional three-year horizon.
To help teams make progress, we started using a framework built around one of Gusto's values: "dream big and make it real." The idea is to develop a concept direction strong enough that the team can plan their iterations from it, and we've pulled in our time horizons, letting the goals drive how long the horizon needs to be. Each iteration should be self-contained and releasable. It solves a specific customer problem and comes with a clear learning plan.
Designers scope their own releases rather than waiting for handoffs. They own the question: "is there a simpler solution that builds toward my concept?" Each phase ships as a whole experience, not just the functional layer of something bigger.
I'm running this with multiple teams right now, all at different stages. On one, I built a prototype to show what the increments were building toward. On another, we're using the vision to scope iterations directly.
The Seed
Every time someone dumps context into an LLM, the truth splinters a little. You end up with artifacts nobody reads and a team working from different versions of reality.
The seed is a new coordination primitive for teams that use AI tools. It's a structured markdown file: what the team knows to be true, what we think but haven't proven, and what we still need to decide. It comes before the PRD, before the brief. It's the starting point the whole team generates from.
It's written to be parsed by AI tools, not read linearly. Anyone on the team can drop it into their tool of choice and pull exactly what they need. A summary, a decision list, a catch-up for someone coming back cold. The seed doesn't change. The artifact changes based on who needs what.
I designed the seed to test a hypothesis: if you give a team one structured doc that AI tools can actually parse, does the problem go away? Does everyone stop working from a slightly different version of what's true? (And does that one really entrenched debate finally unstick?)