What We Heard

Smart Cookies, September 23, 2026, Seattle

Most AI conversations fall into one of two camps: it’s the future, or it’s the end of the world. On September 23, we set both aside and spent the evening on the part in between: what actually happens when you roll AI out across an organization.

Four leaders joined us, each from a different part of the business:

• Kylie Alfano, Senior VP & COO, Tableau
• Tom Van Buskirk, VP of Technology & Engineering, PitchBook
• Andy Stetzler, AI Enablement, Gates Foundation
• Lee Peterson, Deputy CISO, Nordstrom

Bryon Scharenberg moderated.

Key takeaways

1. Access is the easy part.
The Gates Foundation gave every employee access at once and went from zero to about 95% adoption in three months. Then usage leveled off. At Salesforce, a Slackbot agent spread across the company, while an early Claude rollout faded once people found the results were wrong. The difference was whether the tool actually worked in people’s daily work.

“When you step back and ask why, it’s because it worked.” (Kylie Alfano)

2. Data comes first.
Every panelist came back to it. AI built on messy or ungrounded data produces results people stop trusting.

“AI is only as good as the data you put into it.” (Andy Stetzler)

3. Security works best as a path, not a wall.
Nordstrom spends against the risk it actually sees, not the risk the industry talks about. It also built a security chatbot so teams can check an idea against company policy before they build.

“There’s no shutting it down. We need to understand how to enable it securely.” (Lee Peterson)

4. Follow the people closest to the work.
At the Gates Foundation, the top 1–2% of users account for a large majority of token use. At PitchBook, the best ideas come from the people most hands-on, and they’re turned into shared standards for everyone.

“It’s incredible how much AI amplifies the expertise of experts.” (Tom Van Buskirk)

5. Measuring outcomes is still unsolved.
Costs, usage and deployments are easy to track. Proving customer value or better decisions is much harder, and the panel said so plainly.

“If someone up here can tell me how to measure outcomes, I’d absolutely love that.” (Tom Van Buskirk)

“Our biggest question is how we use AI to make better decisions.” (Andy Stetzler)

6. Reskilling, not replacing.
The panelists described jobs changing shape, not disappearing.

“Six months ago, everyone said the analyst was dead. My analysts are busier than ever.” (Kylie Alfano)

From the audience

Asked how to tell real adoption from people just going along with it, Kylie was direct: “People should want to use it, and it should give them value.” Andy said the goal isn’t to push everyone. It’s to lift the middle of the adoption curve. And asked about their biggest misstep, PitchBook admitted it had overestimated what an LLM could do before investing in ground truth and evaluation.

Andy summed up the mood of the room with one word: “I’m an apocaloptimist.”

Who else needs to be in the room?

Bryon closed the discussion by asking each panelist what they need from outside their own function. Andy said HR, to keep people at the center as teams become part human, part AI. Lee said early communication with the business. Kylie and Tom both said a strong data foundation, and Tom added clear communication with executives.

It’s a question worth taking back to your own team.

Thank you to our panelists for being so candid, and to everyone who joined us and brought questions to the room. If you were there, you’re officially in the cookie jar. 🍪

If you’re working through these decisions in your organization, we’d be glad to continue the conversation with you. Connect with KG Digital