Day 1 was about possibilities. Day 2 was about the walk home.
The Shift
The room was filled with anticipation and excitement. Everyone was looking forward to defining the how.
Day 2 delivered on its promise: this was about how you actually make it happen when you get back. Less inspiration, more implementation. The activities were designed to take the use cases we’d selected and turn them into something you could hand to someone on Monday.
The Deep Dive Canvas
The first major activity was the Deep Dive Canvas — a structured look at not just how an agent works, but how it would work specifically within your environment. The prompt: choose one use case and go deep on the how.
The Kwik Trip team chose Care Navigator. I wanted One Support Agent to Rule Them All — it’s the most futuristic use case we had, the most complex, and I felt like working through it would force the hardest thinking about what KT would actually need to make it real. The team went a different direction — Care Navigator, which happened to be my second choice.
Care Navigator is about helping coworkers find quality healthcare at the best price. The user need on the canvas: finding quality healthcare at the best price.
The canvas pushed past the “what” into the mechanics, and here’s what the KT team mapped out:
Orchestration Pattern: Pattern 2 and 3 — human-led agents operating semi-independently, with humans governing and monitoring
How coworkers interact: Primarily chat (Teams, web, or AI app), with the door open for integration into existing KT systems like Bamboo and SAP
What the agent needs to know:
- Health plan information
- Benefits eligibility
- Doctor and provider directory
- Routing logic
Where that data lives today: Customer data services, SAP, historical records
How it collaborates: Coordinates with other agents via sequential hand-off — with a specific concern noted about ensuring coworkers don’t get lost in an orchestrator maze
What success looks like: Measuring frequency, accuracy, and failure rates; tracking healthcare cost outcomes and coworker satisfaction; dashboard visibility
What triggers it: Human request, with event-based triggers possible when benefits data changes
The discussion also pushed into what a fully autonomous version of Care Navigator could look like. Instead of waiting for a coworker to ask a question, the agent would review medical claims as they come in, analyze whether the coworker used a provider that delivers quality care at the best price, and automatically issue the incentive for making that choice. No form to fill out, no process to navigate — the system sees the good decision and rewards it in real time. That’s Pattern 3 in action.
Documenting All Six
After the deep dive, the team worked through all six use cases using the AI Acceleration Plan reference — a structured set of questions across four areas:
Use Case
- Title and description
- Which orchestration pattern does it use?
- What are the key risks and gaps that need to be addressed for it to succeed?
Impact
- What specific business outcomes are you aiming to achieve?
- How will you measure success? Which metrics define progress?
Environment
- What data foundation or integration needs to be in place?
- What tech stack is required?
Action Items
- What are the critical next steps to complete immediately?
- What would you like Microsoft to provide to help implement it?
What came out of working through all six: the team kept wanting to go deep on the roadblocks and challenges. Less about the vision, more about what’s actually in the way. But we walked away with something concrete — the details we need to actually get moving on all six ideas.
One Small Step
The closing activity was deliberately small. The instruction: agree on a quick win you can land within your organization starting next week. Four P’s to frame it:
- People — Actions that involve your team or your C-Suite
- Process — Processes you can act on right away
- Policy — Policies that can be quickly implemented
- Platform — Products or tools you use
For Kwik Trip, the commitments landed in People and Process.
People
- Review the use cases and debrief with the core team
- Explore the interest of these agents with the business
Process
- Define our prioritization process
- Identify what current processes we can incorporate this into
- Decide if we should have office hours and how they should be structured
What Stuck With Me
The support agent conversation was the one that kept coming back to me. People still want to call in — they’re not going to stop. And in a lot of organizations, they’re being pushed to use a chatbot whether they like it or not. But what if the chatbot didn’t try to replace the call — it just redirected it smarter? Chat with a human, human hands off to AI. The handoff chain matters as much as the technology.
The other question that didn’t get fully answered: can frontline workers use these tools? Not managers, not IT — the people actually doing the work. That’s the open door question, and it’s the one worth pushing on.
What Comes Next
- June 3 — Virtual Reunion: reconnect with the group, share lessons learned, hear from experts on topics you requested
- Next two weeks — 1:1 post-workshop debrief with Microsoft: understand what excited you most, uncover key challenges, identify what support is needed to advance your use cases
Related Posts
Previously published:
- Building an AI Usage Policy Your Team Will Actually Follow — The Green-Yellow-Red framework for a one-page policy that provides real guidance
Previously published:
- What Happens When AI Stops Being a Lecture — AI Experience Accelerator Day 1 — The format, the frameworks, and the use cases Kwik Trip walked away with
- Using AI for Difficult Stakeholder Conversations — When you know what you need to say but can’t figure out how to say it without blowing up the relationship