Recorded as I walked along the beach at sunset. Excuse the seagulls.
The end of middle management is closer than most leaders realize — and what comes next is more interesting than the headlines suggest.
In January 2020, just when COVID became a thing and a year before anyone outside a research lab was talking about LLMs, I wrote a piece for Training Industry called The Future of Work: Driverless Teams. I imagined a workplace assistant called Nav. Nav was a chatbot, always on and always ready to help. It could manage your calendar, write emails, assign tasks to colleagues, book you a company-paid Uber to your favorite cafe, and document post-meeting actions while you focused on the conversation. It could also monitor your heart rate variability through wearable device integration. Above Nav sat the Hub — an executive dashboard showing team productivity scores, collective biometrics, and a 72% probability that the recent dip in performance was a rhinovirus (the common cold) rather than Ben’s birthday drinks.
I closed that piece by suggesting managers would shift toward being coaches, advisers, and mentors. The middle-management layer, I argued, was about to be quietly disassembled by software.
Six years later, some of it has happened. Nav exists in pieces, scattered across Claude, ChatGPT, Replit, Notion, Slack, and a dozen biometric wearables. The Hub exists in pieces, too. The executive dashboards I described are real, sold by half a dozen vendors, and being adopted faster than anyone predicted.
So here’s the question that pulled me out for this morning’s walk: if the last six years moved that fast, what does 2031 actually look like?
I want to give you my honest answer. I’m an optimist on most of this. But there are things to flag.
What changed in the last five years
Just five years ago, we were in the thick of COVID. In New Zealand, Auckland was blockaded from the rest of the country, with only essential services allowed to pass through. Every organization that could went remote. Managers and senior leaders suffered the most — at least from the feedback I received. They liked wandering around the office, being seen and heard. When you can’t physically lead your team, something is lost.
As the world reopened, many people discovered they preferred remote work. Hybrid is now normalized. But the bigger story — the one nobody in 2021 saw coming — was the appearance of useful AI. The chatbots that have infiltrated our lives have made 2026 a confusing, exciting place to be.
Here’s an observation I keep coming back to. When the internet first became usable through browsers like Netscape, the goal was visual beauty. I bought a book called Learn HTML in 24 Hours and built web pages that, by today’s standards, looked like a clown’s birthday card. I was ecstatic. But Cross-browser compatibility was a nightmare. Then came Chrome, then mobile-first design, then a slow stripping back of ornament.
Now we sit, conversing with LLMs via what someone from 2003 would consider a command-line interface. Text. One large input box. Who predicted that? Nobody. And yet it was always coming, because what we actually want from a computer is to talk to it.
I now ship production apps by talking to my computer. I use WisprFlow to transcribe my voice and brief whichever LLM I’m working with. I cross-reference outputs from Claude, ChatGPT, and Gemini — they get cranky with each other sometimes, and as the human operator, I’m happy to take the slack.
If we leapt from Zapier in 2021 to armies of intelligent agents in 2026, where will we be in 2031?
The numbers everyone’s working with
Before I go further, the data. Three forecasts that I think every leader should have in their head.
McKinsey projects that by 2030, up to 30% of current hours worked in the United States could be automated, accelerated by generative AI, with around 12 million occupational transitions required by the end of the decade. The bottom 40% of wage earners are up to 14 times more likely to need to change jobs than the top.
The World Economic Forum’s Future of Jobs Report 2025 — surveying over 1,000 employers representing 14 million workers — projects 92 million jobs displaced and 170 million new jobs created by 2030, a net increase of 78 million. They expect 39% of core workforce skills to be transformed or obsolete in that window. The fastest-growing skills are AI and big data, networks and cybersecurity, technological literacy, creative thinking, resilience, flexibility and agility, curiosity and lifelong learning, leadership and social influence (this is what FLAME is all about).
Goldman Sachs projects data center power demand to accelerate 175% by 2030 from 2023 levels. Which is its own story, and we’ll get to it.
You can take or leave the institutions producing these numbers. I cite them not because Davos and McKinsey are oracles — they aren’t — but because the directional consensus across very different methodologies is striking. Something genuinely large is moving.
The new architecture
Here’s the structural shift I think most futurists are missing. The org of 2031 isn’t going to be a flatter version of the org of 2026. It’s going to be three layers stacked together, and each layer answers a different question.
The base layer becomes intelligence. AI is the new electricity — always on, ambient, doing the cognitive work that used to be distributed across managers, analysts, coordinators, and ops people. Every workplace will have more cameras, more microphones, more sensors quietly ingesting input and surfacing recommendations. A doctor sees a patient while an always-on assistant interprets, records, suggests, and writes the prescription to the file. A teacher runs a classroom while the AI flags the student who’s struggling. A bank customer FaceTimes an avatar that confirms identity through biometrics, handles the inquiry, and offers a human if you want one. The administrative burden — the thing that has been crushing professionals for thirty years — finally lifts.
The middle layer becomes coaching. This is where the management story breaks from the standard takes. The standard story is AI eliminates middle managers. I think the truer story is AI eliminates coordination, which is what middle managers were doing, and the role transforms into something better. When AI handles assignment, tracking, status reporting, and performance measurement, what’s left for the human in the middle is the irreducibly human stuff. Holding a team through uncertainty. Helping a struggling person find their footing. Naming what’s actually going on in a relationship. Building belief. Regulating the emotional weather of a group. That’s coaching. The skills McKinsey and the WEF both name as fastest-growing — resilience, leadership, social influence, and agility — are coaching skills (the kind I offer at Values Institute), not management skills.
The top layer becomes storytelling. When AI handles execution, the constraint on senior leaders moves from can we do this to should we, and what for. The most valuable thing a CEO does in 2031 is name what matters and why. Narrative. Meaning. Direction. The values work that used to feel optional becomes the entire job.
This three-layer model also explains who keeps a job and who doesn’t. The roles that survive are the ones that touch one of these three layers. The roles that don’t are the coordination and admin roles in between, which is exactly the band the data points to.
What humans are actually for
Some specifics worth flagging.
You’ll still want a human real estate agent — I don’t want a robot showing me around a house. Nurses, because most people are far more comfortable with a human than a robot (for now). Coaches, speakers, school teachers, doctors freed from admin, firefighters accompanied by firefighting bots, actors and performers, streamers (we are fascinated by other people), front-of-house staff in hospitality.
What changes is what happens behind the front desk. Receptionists stay; the back office automates. The supermarket cashier is gone, but the boutique that pitches itself as a deliberately human experience charges a premium for it. McKinsey’s data is blunt on this: customer service and office support roles contract sharply. Healthcare is the dominant growth engine — the U.S. alone is projected to need 3.5 million more health aides, technicians, and wellness workers, plus another two million healthcare professionals, by 2030. The care economy explodes, partly because of aging populations and partly because it’s the sector where humans matter most.
Three categories of worker emerge inside organizations. Those who prompt and train the AI and the robots. Those who work exclusively with humans. And those who look after the humans — and increasingly, the robots — within the workforce. Most of us will be one or another of these.
The energy fork nobody’s pricing in
Here’s the thing almost no future-of-work piece is touching, and it might be the single biggest variable in the whole forecast.
Every prediction in this article assumes always-on AI is cheap and abundant by 2031. That assumption is the actual battleground of the next five years. AI data centers are projected to consume 945 terawatt-hours annually by 2030 — equivalent to the entire electricity consumption of Japan — and there are three competing answers to where that power comes from.
The first is space. SpaceX filed plans with the FCC in January 2026 for up to one million solar-powered satellites that would serve as orbital data centers. Musk’s pitch: you’re power-constrained on Earth; in space it’s always sunny. Google has its own version called Project Suncatcher. Blue Origin announced TeraWave. China filed for a 200,000-satellite constellation. Google’s own feasibility study projects this could be cost-effective relative to ground-based data centers around 2035 — note the date — if SpaceX’s Starship scales to 180 launches per year by then.
The second is nuclear. Tech giants have committed over $10 billion to small modular reactors, with 22 gigawatts of projects in development globally. Microsoft is restarting Three Mile Island. Meta signed a 20-year deal with Constellation. Google is buying from Kairos Power. Amazon and Oracle are in. The first commercial SMR-powered data centers are expected online by 2030.
The third is the unsexy answer: grid expansion and natural gas peakers. Most of the near-term demand will be met this way, which is why fossil-fuel investors are quietly bullish on the AI revolution.
This matters because there are two possible 2031s, and they look very different. In the first, compute is abundant, agents become as common as electricity, and most of what I’ve described unfolds on schedule. In the second, power constraints bite, AI access tiers by ability to pay, and the agent revolution arrives unevenly — wealthy economies and wealthy employees get the full Nav-style assistant, everyone else gets a stripped-down version. The inequality story sharpens.
I lean toward the first. But I’d rather flag the fork than pretend it isn’t there.
The dividend: what we get to do instead
Here’s the part I’m most excited about, and the part almost nobody is writing.
When AI absorbs the cognitive grunt work and robotics absorbs the physical grunt work, what we actually get back is human bandwidth. Attention. Capacity. Hours that used to be eaten by compliance, admin, and coordination. The interesting question isn’t what jobs disappear. It’s what we choose to redirect that recovered capacity toward.
Some of it is already showing up. The Ocean Cleanup has removed over 15 million kilograms of plastic from oceans and rivers as of 2026, with AI guiding every deployment by predicting where plastic will concentrate based on currents, wind, and historical accumulation. AI-driven routing has improved their collection efficiency by 68% over traditional methods. Global Fishing Watch, a partnership between Google, Oceana, and SkyTruth, monitors over 65,000 fishing vessels in near real time and has revealed that an estimated 20% of the world’s catch comes from illegal fishing — Indonesia alone has used the data to crack down on hundreds of vessels.
Closer to home, Sustainable Coastlines uses AI to identify the sources and patterns of coastal pollution and has removed enough trash from New Zealand and Pacific shorelines to fill nearly 45 shipping containers.
This is the version of the AI story that doesn’t get told enough. By 2031, I think we’ll have made measurable progress on problems we’ve been losing on for thirty years. Ocean plastic shrinking rather than growing. Illegal fishing exposed in real time. Rewilding accelerated by AI-modeled habitat planning, drone seeding, and bioacoustic monitoring of biodiversity recovery. Land-use shifts that nobody’s quite forecasting yet — the combination of GLP-1 drugs reducing meat consumption, lab-grown protein scaling, and AI-optimized agriculture could quietly hand back vast tracts of pasture to the wild.
The same systems absorbing our admin work are giving us back the bandwidth to fix what we broke. That’s the optimistic 2031, and I think it’s the more likely one.
The thing to flag
I want to end with the thing that’s been bothering me, because if I don’t name it, I’m just another voice cheering the new architecture.
I was listening recently to a Y Combinator talk where an expert was describing how her company had moved everything into closed loops. Every action captured. Every commitment tracked. Every gap measured and fed back into the next sprint. Her engineers were 10x’ing their output. The numbers were real. And it sounded, honestly, kind of horrifying.
That reaction is worth taking seriously. The closed-loop org is the natural endpoint of the architecture I’ve just spent 2,000 words describing. AI as the base intelligence layer means every conversation captured, every signal logged, every loop closed. From one angle, it’s the dream — finally, the org becomes a continuously self-correcting system rather than a series of disconnected episodes. From another angle, it’s a place with no outside. No room where you’re not being optimized. No half-formed thought that doesn’t get tracked. No wrong turn the system doesn’t quietly correct. No private margin for the unmeasured human moment.
Some loops should absolutely close. Compliance loops. Quality loops. Coordination loops. The admin burden I keep talking about — close it, kill it, never look back. Doctors freed from documentation. Teachers freed from grading. Engineers freed from boilerplate. The hour back at the end of every working day. Genuinely good.
But some loops have to stay open. Creative loops. Recovery loops. Relational loops. The conversation between two people that doesn’t get logged. The walk on the beach where an idea forms in private. The team meeting where someone is allowed to be wrong and confused without it being captured in the post-meeting summary. The space where a values question is held rather than resolved.
This is, I think, the actual leadership question of the next decade. Not should we adopt AI — that’s already answered. Not should we close the loops — they’re already closing. The question is which loops do we deliberately, defiantly, keep open?
A manager closes loops. A coach holds them open long enough for something to develop that wouldn’t survive measurement. The companies that win the next decade won’t be the ones with the tightest feedback systems. They’ll be the ones whose leaders had the wisdom to know which gaps were worth keeping.
That’s a values question. Which is the question this whole century turns out to be about.
I’ll see you in 2031. Probably out the back at the Mount, waiting for the next set.