Are AI Agents coming for middle managers?
Relax, they only want your status reports… and maybe your job
Welcome to FullStack HR, and an extra welcome to the 19 people who have signed up since last edition.
If you haven’t yet subscribed, join the 9300+ smart, curious and like-minded HR people by subscribing here:
Happy Thursday!
The last few weeks of spring, just before summer begins, have brought some intense weeks!
Two major highlight the last couple of weeks.
First one was educating 140 people simultaneously last week on how to harness the power of generative AI.
The energy in the room was insane, it was so much fun!
Thanks a lot Helsingborgs-familjen for spending that time with me.
Second was educating and tinkering with AI strategies together with 50 (!) CEO:s.
In both cases, we experimented with the technology, solved real problems, and it’s fun seeing people immediately picking up AI skills and starting to resolving challenges with AI.
My calendar is more or less fully booked up until summer but I do have a couple of spots left in August and September, so if you need someone to kick things off after summer to infuse your team with AI-knowledge - reach out!
Fullstack HR is brought to you by…OneLab - AI-Powered insights for better workplace health.
Creating healthy and health-promoting workplaces is not only good for the individual, it also strengthens the company and society. OneLab helps you take the next step in that work – with the power of AI.
OneLab’s platform provides HR and managers with intelligent insights that make a difference. OneLab's AI assistant analyzes data from health examinations, employee surveys, and sick leave, and provides tailored recommendations based on your needs.
It has full knowledge of the ‘Arbetsmiljöverkets författningssamling’ and is always available to support your work environment and health efforts!
Every health examination in our platform becomes an anonymized, data-driven basis for decision-making that helps you identify health risks in time and prevent ill health before it arises.
Companies like Deloitte, Nasdaq, Electrolux, Länsförsäkringar, and Warner Bros already trust OneLab – you should too.
Listen to the article - powered by ElevenLabs - try it for free here. (It’s like black magic.)
I have spent most of my career in the middle seat as a middle manager. Even when acting as CHRO, I would argue that you have had me parked in that in-between zone, catching projects from the top floor and translating them for the people who make things happen.
My path to HR even began in a very classic middle-manager job: running a ski school where every equipment mix-up or blue-run mishap somehow landed on my desk.
Middle management is all about coordination. You keep people aligned, pass information along, and make sure everyone knows what they should work on. In other words, you act as mediator between the top brass and the rest of the organisation.
Is that not the perfect role for an AI agent to take over?
Since just before Christmas I have been tinkering with agents. I put “agents” in quotation marks because there is no single definition, but here is mine: an agent is an AI you hand a task to and it solves the job on its own, whether that means calling a candidate or drafting a project plan for an organisational change.
My first goal was to build agents that handle one specific task. It is actually quite easy to create a voice agent that automatically calls every new applicant. (Reach out and I will show you how.) Yet the more I build, the more I realise that agents truly shine at orchestration.
Here is how it works.
When building agentic workflows, everything starts with a trigger: an application arrives, an email goes out, a meeting wraps up. The trigger wakes an orchestration agent. This agent has a small tool belt and one clear job: pick the right tool for what just happened. If the trigger is an application, the agent might choose the “extract information from PDF” tool. If it is an email, it might pick “read this mail” plus “research this person.”
The orchestration agent delegates tasks, tracks progress, and loops the results back to wherever they belong. Sounds familiar, right? That is exactly what a middle manager does.
Imagine a strategic lead-team meeting. We record it, run a transcript through the orchestration agent, and tasks flow automatically to the right teams, people, or other AIs. And then the orchestration agent loops information back to the people in the meeting that the task has now been executed.
Technically, this is already possible. (and I believe it can be made EU AI Act-compliant.)
How far off is this reality? The tech is here today. Whether organisations are ready is another story. Research from i4cp points that organisation in general are not ready to even adopt agents in the workforce, let alone then
And even if they are, do we truly want to replace human middle managers?
That is the big question.
Does these “manager agents” have feelings? No, but notice how much context it can absorb. Connect it to Slack and it picks up the organisation’s pulse in almost scary detail. But then another concern arises - privacy boundaries. Where do we draw the line between helpful context and unwanted surveillance? So once again, all of this is not easy, even if the tech is pretty straight forward.
Many people raise empathy as the final barrier. Right now the agent only mimics it, but those simulations get better every month. This is also both a blessing and a curse. A curse because even if we connect it to all our systems, will it truly meet employees where they are at? Will the mimicked empathy be enough for us humans in the organisation?
But it can also blessing because your manager agent won’t have a bad day.
And a 2023 McKinsey survey of 700 middle managers found they devote nearly one full day each week to pure admin and spend roughly seventy-four percent of their time on tasks they call low value, leaving less than a third for coaching and strategy, little wonder they are the most burned-out cohort in the org chart. And couldn’t then an orchestration agent solve that while an empathic coaching agent could cater for the employees?
Middle management was invented to solve human limits: attention, memory, context. Software does not share those limits. The moment AI routes work faster and with perfect recall, the rationale for that layer starts to crumble. That is what every digital disruption teaches us: once a constraint disappears, the structure built around it follows. So the smart play is not to defend the middle managers but to move up a row. Own the principles the agent optimises for, curate the data it sees, audit the outcomes it produces. Judgment stays human, just at a higher lever point and vastly more scalable.
Most likely we’ll see a mix in organisations, a mix of AI agents and humans and I foresee that we will see org charts looking like this in the future.
But still, the chart above sketches one possible destination, not a final blueprint. What matters now is how we travel there. Give routine coordination to the agent, then use the time you win back to do the work software cannot: set direction, nurture trust, and decide when the algorithm is wrong. Pilot small, low‑risk workflows, measure the lift, and share the results. Each experiment is a step in redesigning the org around judgment rather than traffic control.
While I’ve been arguing about that middle management might disappear here, I don’t think that will happen over night. Middle management is not being deleted for now, but I do believe it is being distilled. The role shrinks in volume but grows in value. Those who learn to pair their own judgment with agent horsepower will shape the next org chart.
That future is already knocking.
Let’s open the door.
Agreed. The middle layers with be thinning out (and replaced by agents) but they won't disappear completely. There's still plenty of people stuff to do.