Building the workforce for agentic AI: Skills, teams, and what leaders need now

Most organizations are still figuring out their platform strategy. Executive interest in AI is high, but translating that interest into committed buy-in with real budget behind it remains a work in progress for many teams.

Even where those pieces are coming together, the question that gets less attention is whether the people building and operating these systems have the skills, the mindset, and the training to deliver on what agentic AI actually requires.

And the distance between having a capable platform and having a capable team is where most agentic ambitions will quietly stall.

In this episode of the Agentic Edge podcast, hosts Micah Smith and Kate Ressler sit down with Mike Reynolds, Business Technology Executive - Service Digitization at KeyBank. He oversees automation platforms, including Automation Anywhere, enterprise low-code with OutSystems, and the architecture and infrastructure teams that tie it all together. With over 20 years of delivering technical solutions in financial services, he brings a practitioner's perspective on what it actually takes to build, retool, and lead teams through the shift from task-based automation to goal-driven agentic AI.

👉 Watch the full episode on YouTube or listen on Spotify / Apple Podcasts.

What follows in this post are the ideas from this conversation worth sitting with.

The people who know the process are the ones you want building agents

There's a common instinct when new technology arrives: go hire for the technical skill. Mike took a different path early on with robotic process automation (RPA), choosing to train people from the line of business rather than bringing in developers or engineers from the outside.

That team could challenge assumptions about how work actually flowed, trace processes upstream and downstream, and build trust with business partners because they genuinely understood the pain.

You can teach technical skills. Replicating the process intuition someone builds over years of doing the work is much harder. That foundation is now proving to be the natural launchpad for agentic AI.

People who became strong automation builders also tend to understand where agents can add value and where they can't, because they already think in terms of decision points, exception paths, and where processes break down. The talent pipeline that started in the RPA era created cross-functional fluency that agentic systems demand.

Ideas that were rejected years ago are now viable

Mike has been tracking automation ideas since 2019. In the early years, volume was steady at four or five a month. Today it's averaging 20.

In addition to new ideas, his team has reevaluated hundreds of previously rejected ideas. Agentic AI has changed the equation on many of them. Processes that were too complex, too dependent on judgment, or too variable for traditional RPA are now within reach. What was a clear "no" three years ago is becoming a credible "yes."

At the same time, more people across the organization are getting involved in ways that simply weren't happening before. Executives are vibe coding demo sites overnight. Business partners are actively requesting access to sandbox environments they previously had little desire to enter.

The rules of yesterday are blending into something new. Figuring out how to capture that innovation energy and channel the best ideas toward production is now a core leadership design consideration.

Governance is the infrastructure that earns trust for the next phase

Moving fast inside a heavily regulated industry is more of a design problem than a speed problem. Mike's team built on the foundation that years of RPA delivery established: repeatable processes, appropriate controls, and a track record that risk partners could point to.

That foundation is what makes it possible to move forward with agentic AI in a way that satisfies governance requirements while still capturing real value.

The more nuanced design consideration is getting people comfortable with AI outputs that carry a level of variability, deterministic automation does not. People across the organization understand AI at different levels, and bringing those levels together so everyone can evaluate testing results and assess agent behavior is an active effort.

Every day, more tools emerge to help with guardrails, and that progress is making the path from pilot to production clearer.

Most teams underappreciate one practical point: RPA and agentic AI work well together rather than in competition. Some processes call for the repeatability and cost profile of a bot doing the same thing the same way every time.

Others call for the reasoning and adaptability of an agent. And some call for a combination, where an agent coordinates with bots through something like Model Context Protocol (MCP) to get the best of both. The teams thinking clearly about that spectrum will manage cost and complexity better than those treating it as a binary choice.

Whatever you learn today, be willing to relearn tomorrow

When the question turns to which skills matter most, Mike leads with problem solving. Across more than two decades and multiple technology shifts, from Pascal to R to whatever comes next, the constant has been the ability to understand a business problem, think through it clearly, and figure out how to solve it.

He describes a layered approach to upskilling. There's a prescriptive side: training on evaluation metrics, observability, prompt drift detection, and the mechanics of testing whether an AI system is producing the results you expect. These are areas most agentic training programs don't cover well enough.

Then there's the curiosity side. Every week, the team gets dedicated time to learn whatever they want, whether that's a new feature in the console, a business process deep-dive, or a tutorial on something that caught their attention. He combines structured paths with open exploration deliberately.

What stands out is the emphasis on closing the gap between technical and business fluency. Technical people benefit from understanding the business well enough to articulate why their work matters, and business-minded people benefit from staying close enough to the technology to know what's possible.

Investing in people is how capability becomes durable

There's immense pressure right now to deliver valuable results on short timelines. That pressure makes it tempting to deprioritize learning. But, what if you don't invest in your people and they stay at the same level?

In a world where AI capabilities are shifting weekly, standing still carries its own compounding cost. Bringing in contractors with a specific skill set can bridge some gaps in the short term, accelerating the team and transferring knowledge. Long-term capability has to live on the team.

That means making deliberate decisions for each investment: is this a personal development goal or a team-level priority? How much time should the team spend? And training goes beyond technical skills. Understanding business processes deeply reinforces the ability to design effective solutions, and the two compound together.

The organizations building durable capability in their people, alongside their platforms, are the ones that will replicate success across use cases and scale. A platform is only as strong as the team that knows how to fully leverage it.

Go Deeper

This conversation covers more ground than what's captured here. The quickfire section toward the end is particularly worth watching, with direct takes on vibe coding in the enterprise, AI governance as both enabler and friction, and the tsunami of ideas that can paralyze teams. Useful for anyone leading automation programs right now.

👉 Watch on YouTube | Subscribe on Spotify | Subscribe on Apple Podcasts