"How can we actually reward and recognize people using this in an impactful way — innovative and solving a real problem?" — Chris Olson, Scout OS

Organizations are investing in AI capabilities. Employees still aren't using them in ways that move the business forward. The real work people leaders brought into this live in-person exchange was about activation — building recognition and incentive structures that tie AI experimentation to direct business outcomes, close the gap between tool availability and frontline adoption, and reward smart experimentation over empty activity.

The harder problem underneath: most organizations don't yet have an AI adoption policy. And the ones that do have policies that are very, very broad.

Leaders in the conversation

Anton Maletich (Alter Domus), AI Practitioner, arrived with field experience — a multi-agent workflow built from scratch to automate SuccessFactors training reports, complete with the failure that taught him more than the build did.

Angie Zeigler , Strategist, anchored the room in the business challenge and brought the insightsoftware Shark Tank case study as a tested model for what recognition-through-competition actually produces.

Chris Olson (Scout OS), AI Advisor, ran a live demonstration of agent architecture, permission design, and how organizations can — and can't — control what their people build.

The rest of the room filled in the signal: practitioners working through the same adoption stall, trading notes on what's worked and what hasn't.

Moving Past Poster Recognition

The room landed quickly on a shared frustration: most AI recognition programs reward presence, not performance.

If your program awards the employee who logs the most AI usage, you've already lost. As Chris Olson put it:

"Think of it like turning on the tap and letting it run. You can do that with agents." — Chris Olson, Scout OS

A motivated employee can run the largest model on meaningless queries all day and inflate their usage numbers without producing anything. Token volume is not an adoption signal. It's a gaming invitation.

The shift the room was working toward: recognition tied to operational outcomes — cycle time, output quality, error rate, decision speed — with those metrics visible to employees before the program launches, not after the first quarter's data comes back wrong.

Chris named where people leaders fit:

"That's why I focus on the human resource aspect of this — they're basically going to be the wedge, the key that drives the deployment of AI in organizations." — Chris Olson, Scout OS

Recognition isn't a bolt-on reward. It's the infrastructure. People leaders are the ones who build it.

The Internal Hackathon Engine

The most fully formed model in the room came from insightsoftware, shared by Angie Zeigler on behalf of Laura Jones, who designed and ran the program.

A companywide AI hackathon built on a Shark Tank format. Employees developed new product or workflow ideas using AI as a core part of the process — not just a tool. Teams submitted using a PR/FAQ format: a press release and FAQ that forced them to articulate the problem, the solution, and the value before building anything.

The judging panel was the CEO, a board member, and the CTO. Finalists competed in multiple rounds, with prizes of approximately $1,500, $750, and $500 for the top three.

But the prizes weren't what moved people.

Visibility to senior leadership was the activation mechanism. Pitching AI-built work to the CEO and CTO changed what experimentation meant organizationally — it became a career move, not a compliance exercise. Multiple participants were promoted. One employee's idea went into production as a new product.

"It was really a whole product management process that they used AI to develop." — Angie Zeigler

The program now runs multiple times per year — a repeatable pipeline for surfacing localized ideas, stress-testing them with executive scrutiny, and converting the strongest into enterprise-level outcomes.

Designing Cheat-Proof Frameworks

The breakout groups didn't just generate recognition ideas. They immediately pressure-tested them.

One group asked AI to design a recognition program — then asked it to map every way an employee could beat the system. The output was more useful than the original design prompts.

"You're just recognizing that I'm busy — not that I'm working towards something that's going to provide value for the organization." — Community Voice, Breakout Group

Employees know when they're being seen versus when they're being measured. The difference matters.

On the technical side, Chris Olson walked through the architecture required to prevent unchecked agent access. The answer is user-scoped permissions: an agent should only act as the user who built it, mirroring that user's existing system access. Without this, sharing an agent across a team creates unintended access escalation.

Model Context Protocol (MCP) servers are the connective tissue — allowing agents to communicate with HRIS systems, CRMs, and project management platforms while respecting authentication boundaries. Knowing which MCP servers your organization has enabled is a prerequisite to any agent rollout at scale.

Anton Maletich surfaced the practitioner version of this risk.

"This is one of the blind spots. Make sure you know the capabilities of your system." — Anton Maletich, Alter Domus

He'd spent significant time building an eight-agent workflow to automate SuccessFactors training reports before discovering the platform couldn't accept the core data file. He pivoted to Databricks — which ultimately worked — but the lesson was clear: ask the tool what it can't do before you build around assumptions.

---

What to try next

1. Map the cheat vectors before you finalize the program design. Ask your team or AI: how would a motivated employee game this program? Treat the output as a required design input. Programs that skip this step produce compliance theater in the first quarter and disappointment in the second.
2. Use the PR/FAQ format before anyone builds anything. The insightsoftware hackathon required a press release and FAQ before prototyping — a forcing function that surfaces whether the problem is real and whether the team can articulate value to a non-technical audience. Use it as the entry requirement for any internal AI pitch program.
3. Ask the tool what it can't do before you invest build time. Before any agentic workflow project, explicitly ask the model what limitations or constraints apply to your use case. This surfaces blockers that won't appear in product documentation — and prevents the failure mode Anton described.