"Curiosity, creativity and improv. That's it. That is literally all you need."

People leaders navigating AI transformation aren't lacking tools — they're lacking a practice. That was the live, unscripted signal that surfaced when Mike Hruska (BARYONS) and Marty Murrillo (Precisely) brought a small-circle conversation into a larger room at the ELE Spring Conference 2026: the real work isn't getting teams access to AI, it's shifting from access to active, continuous practice — while keeping human judgment at the center and stopping organizational bias from scaling silently into every automated decision.

Leaders shaping the work

This live interactive discussion grew out of a deliberate collective intelligence experiment — a small facilitated group thought out loud together, captured it, fed it to AI in real time, and then opened the floor to a wider room to add the next layer. What emerged wasn't a polished framework. It was a working exchange between people leaders, coaching practitioners, and L&D professionals thinking through the same friction from different angles: How do I stay in control of this? Where does my real value live now? And what do I do about the biases already baked into my data?

The first question isn't which tool — it's who's driving

The discussion opened with a model worth keeping.

The centaur keeps the human in the lead — AI as power-multiplier, not director. The reverse centaur flips it: the algorithm manages the human, tells them where to go, tracks their eyes, controls the pace. The room recognized both patterns immediately.

One participant described having to actively reset their AI tool — telling it to update its memory, stop its accumulated assumptions, and behave differently. It worked. But the moment landed as something more than a tech tip.

"No one really knows what they're doing — no matter what they say. We're all improvising together."

That's the actual operating condition. Leaders who perform certainty about AI adoption are spending energy they don't have. The ones moving fastest are treating it like improv: curiosity-first, troop mindset, building as they go.

Context is the real input — not the prompt

One of the sharpest practical moves in the room came from a simple story: 28 executives at a global company, most of whom had never used generative AI before. Generic prompts produced generic output. Then two participants tried something different — they told the AI to treat itself like a 20-year-old intern named Sam, gave it real context about the person's tenure and background, and asked it to ask them questions first.

Output quality transformed. Not because the tool got smarter — because the human input got richer.

The practical implication for people leaders: the quality of what AI returns is a direct function of the organizational and personal context you give it upfront. Thin context, thin output. This isn't a prompt engineering tip — it's a practice design decision.

AI as mirror: the administrative shift and the blind spot problem

Two connected signals ran through the middle of the discussion.

The first: AI is already compressing the time it takes to do the work that buries managers. One leader described processing 360-degree feedback for 14 direct reports in two hours — and finding things human review had consistently missed.

"The non-obvious stuff popped up. It works."

That's not an efficiency story. That's a capability story — AI as forensic mirror, surfacing patterns in your own language, your team's feedback, your organization's blind spots. The leaders in the room who were already doing this weren't using AI to replace judgment. They were using it to see what judgment alone had been missing.

The second signal is harder. When one participant described a hiring algorithm that had silently learned a decade of the organization's own gender bias — promoting patterns that favored men, demoting signals associated with women — the room shifted.

"You have to check for biases because it'll just amplify what you have."

The data you feed AI carries the fingerprint of every assumption your organization has normalized. Scaling that without interrogating it first isn't transformation — it's acceleration of the status quo.

The jerk factor — and why your change model is already outdated

Traditional change management assumes a beginning state, a transition, and a future state you're moving toward. That model is broken.

The discussion introduced a more honest frame: in mathematics, the "jerk" is the rate-of-change of acceleration. Not change itself — the change of how fast change is changing. That's what people leaders are actually living through. And it's why the standard change playbook produces exhaustion instead of traction.

"The future state doesn't exist because it's the next iteration. The next iteration. So no wonder you're tired."

The replacement isn't a better Gantt chart. It's a different operating posture — curiosity, creativity, improv — and the recognition that collective intelligence across a trusted peer group is how you navigate what no individual or organization can map alone.

---

What to try next

1. Run one forensic session on your own work. Feed a recent coaching conversation, team debrief, or performance review into an AI tool. Ask it to surface language patterns, themes you avoided, or signals that repeated without resolution. Use the output as a mirror, not a verdict. Early evidence signal: you find at least one pattern you hadn't consciously named.
2. Front-load context before every significant AI task. Before asking for output, give the tool your role, your specific challenge, your constraints, and your audience. Treat it like briefing a capable but uninformed thinking partner — not issuing a command. Early evidence signal: your first-draft output requires meaningfully less reworking than before.
3. Audit one AI workflow for direction. Pick one place where AI is already part of your work. Ask honestly: am I directing this, or is it directing me? If you can't answer clearly, reset — start a clean context, set explicit parameters, reassert what you want. Early evidence signal: you can describe the decision boundary between your judgment and the tool's output in one sentence.