What L&D needs to build as agentic AI joins the workforce, and why remote teams will feel it first
For the last few years, most organisations have treated AI as an assistant. It drafts the email, summarises the report and suggests the code. Someone checks the output, and if it is wrong, the cost is rework.
That phase is ending. McKinsey’s October 2026 guide for banking technology leaders describes what comes next as the “agentic coworker”: AI that owns a role and an outcome, plans work across several steps, acts inside live systems and hands work on to people and to other agents. The guide is written for chief information officers, and most of it is about architecture and control. Read it as an L&D leader, though, and one detail stands out.
Before an agent is allowed to act, McKinsey argues it needs a defined identity, a named human owner, clear limits on what it may do and touch, an escalation route for when it is unsure, and autonomy that is released in stages as it proves itself.
Now ask how many of your managers were given that much clarity when they stepped into the role.
Throughout this article, resilience is defined as “Springing Forward with Learning” - the ability to move through change and come out of it more capable, rather than trying to return to where we were.
The Job Is Changing Shape
The most important line in the McKinsey guide, for anyone responsible for workforce development, is not about technology. It is the acknowledgement that as agents take on routine execution and coordination, human roles will shift towards judgement, exception handling, relationship management and oversight. The authors are clear that workflows and roles must be redesigned together, and that workforce implications should be planned from the start rather than treated as a downstream consequence.
They describe three stages of human oversight:
Human in the loop: a person approves every action before it happens.
Human on the loop: a person steps in for higher-stakes or low-confidence cases and samples the rest.
Human over the loop: a person monitors performance and owns the outcome while the agent runs routine work end to end.
Look at that from the leader’s side. At each stage they do less of the visible work and keep all of the accountability. McKinsey puts it plainly: “Accountability cannot be outsourced to an algorithm.”
That is a significant psychological shift. Status that once came from expertise and output now has to come from judgement. Certainty that came from doing the work yourself now has to come from trusting a system you cannot fully see. In Resilience Wheel terms, it tests Purpose (what is my value now?), Confidence (can I make the call without doing the work?) and Adaptability (can I let go of how I used to add value?).
At every stage of AI autonomy, leaders do less of the visible work and keep all of the accountability. That is a resilience challenge before it is a technology one.
The Adoption Evidence Is Already In
In an earlier edition of this column, I set out Gallup’s finding that AI return on investment is a leadership problem before it is a technology problem. It bears repeating, because agentic AI raises the stakes considerably.
Gallup’s State of the Global Workplace 2026 report found that only 12% of employees in organisations that have implemented AI strongly agree it has transformed how work gets done. After how well the technology fits existing systems, the strongest predictor of effective use was whether the employee’s manager actively supports it. Fewer than a third of US employees in AI-adopting organisations say their manager does.
All of that relates to copilots. Agents ask far more of the manager: not encouragement to try a tool, but ownership of a digital colleague’s decisions and the redesign of a team’s work around it. Meanwhile, Gallup shows global manager engagement fell from 27% to 22% in a single year. Many organisations are about to give their most depleted layer its most demanding new responsibility.
What Is Left for People Is Harder, Not Easier
There is a comfortable assumption that removing routine work frees people for more rewarding work. Sometimes it will. It depends entirely on how the role is designed.
Recent longitudinal research on boredom, burnout and engagement offers three cautions. Role conflict reliably predicted the poorest wellbeing profiles, and a person who owns an outcome but not the process, answering to both a manager and an automated workflow, is a textbook case of role conflict. Autonomy on its own did not protect people from boredom or burnout; it helped only when paired with learning and stimulation. Engagement and exhaustion were also found to sit in the same person at the same time.
So the “over the loop” role can go two ways. Designed well, it is a judgement-rich job with real learning in it. Designed badly, it is hours of monitoring punctuated by high-stakes exceptions, with little practice in between. That combination produces people who look engaged on the survey and feel exhausted at the end of the week.
Photo Christopher Marsh Unsplash
Remote and Virtual Teams Will Feel It First
L&D professionals have been raising a related challenge: how to keep virtual groups and remote workers engaged. Gallup’s 2026 data shows why this sits at the heart of the AI conversation rather than beside it.
Most engaged: 30% of exclusively remote employees are engaged, against 25% of hybrid workers and 24% of on-site staff in remote-capable roles.
Least likely to thrive: 33% of exclusively remote employees are thriving, against 45% of both hybrid and on-site remote-capable employees.
Heavier days: 30% of remote workers experienced a lot of sadness the previous day, against 20% of on-site remote-capable staff. Loneliness was 24% for both remote and hybrid workers, against 18% on site.
In other words, remote people are doing the work and paying for it personally. An engagement score on its own will tell you all is well.
Now add agents. Much of what holds a distributed team together is incidental: the handoff call, the quick question about a case, the second pair of eyes on a colleague’s draft. These are exactly the coordination tasks agents are designed to absorb. Remove them without replacing them, and you strip out the last informal connection some remote employees have. The work gets faster. The team quietly gets lonelier.
Character Becomes More Visible, Not Less
New research from the Norwegian School of Economics and New York University, currently a working paper, adds a further angle. Across nearly 48,000 people in 40 countries, including the UK, employees were more willing to give up a pay rise to avoid a bad leader than to gain a good one. They were also less forgiving of gaps in character than gaps in competence.
As agents take on more of the competence-heavy work, what remains most visible about a leader is their judgement and their character. How do they respond when an agent gets something wrong? Do they own a poor outcome, or blame the system? Do people feel safe to challenge a confident machine output in front of them? The leader whose default answer is “the system decided” is likely to become the new bad boss.
A Practical Plan for L&D
If the practical intersection of AI and workforce development is on the agenda at your next L&D forum, here is where I would start.
1. Write the human role specification before the agent specification
For every agentic deployment, ask the same questions McKinsey asks of the agent. Who owns the outcome? Which decisions do they make? When do they escalate, and to whom? What support do they need to do it well? If the named human owner has not been prepared for that role, the control exists on paper only.
2. Develop leaders for the loop, not just the tool
Tool training teaches people which buttons to press. Leading in the loop needs judgement under ambiguity, the confidence to overrule a machine, the ability to coach a team through redesigned roles, and enough psychological safety for people to flag errors early. These are resilience capabilities. They are built through practice, reflection and coaching, not a one-off e-learning module.
3. Design connection back in on purpose
Measure thriving alongside engagement, especially in distributed teams. When an agent removes a handoff, replace the human contact it carried with something deliberate: a short weekly case review, a peer support pairing, or a team check-in that covers people as well as progress. This is the Support Network dimension of The Resilience Wheel, rebuilt by design rather than left to chance.
The Cost of Waiting
McKinsey closes with a warning for organisations that skip the foundations: they will either slow low-risk work with excessive checking, or discover that an unsupervised agent has already acted. The human equivalent is just as real. Skip the people foundations and you get managers signing off decisions they do not understand, remote staff who look engaged right up to the day they resign, and an AI investment that never shows up in the numbers.
What Good Looks Like
Get this right and the picture changes. Leaders know where their value lies and are confident making the call. Teams trust each other enough to say “that output looks wrong”. Remote colleagues feel part of something rather than simply logged in to it. Agents run faster, inside standards people actually believe in.
The organisations that gain most from agentic AI will not be the ones with the most agents. They will be the ones whose people can spring forward with learning every time the work changes shape.
Five questions to take into your next AI workforce conversation
1. Who is the named human owner for each agent we deploy, and have they been developed for that role?
2. Which roles are moving towards oversight, and have we redesigned them or simply removed tasks from them?
3. Do we measure thriving as well as engagement, particularly for remote and hybrid teams?
4. Where will people learn to challenge AI output, and is it genuinely safe for them to do so?
5. What informal connection are we about to automate away, and what will replace it?
Preparing your leaders for agentic AI?
Find out where your leaders and teams sit across the seven dimensions of The Resilience Wheel, then talk it through with me.
Book a conversation: zcal.co/russellharvey/intro-call
Resilient Leadership Diagnostic: theresiliencecoach.co.uk/resilient-leadership-diagnostic
Resilient Teams Diagnostic: theresiliencecoach.co.uk/resilient-teams-diagnostic
Email: russell@theresiliencecoach.co.uk
Website: www.theresiliencecoach.co.uk