An undersong is the strain underneath the melody. The drip under the thunder. The meaning you only hear once the brass has finished congratulating itself.
McKinsey surveyed more than ten thousand leaders, and 72 percent said their organizations are not ready for what is coming.
That is the loud song.
It is already traveling: LinkedIn posts, webinars, another round of "transformation" decks. Kelly Brogdon Geyer heard it correctly when she said the problem is not awareness. Leaders know. The gap is execution, capability, and people who stay long enough to see whether anything actually changed.
The undersong is quieter.
The company, as we inherited it, was a coordination technology. So was the office. Both were built for a world where intelligence lived in heads, information moved slowly, and the cheapest way to share context was to put bodies in the same building five days a week.
That world is ending unevenly.
The forms have not caught up.
The 72 percent is not a weather report. It is the sound of an old instrument still being played after the key has changed.
What the survey is actually saying
Two-thirds of executives say their organizations are overly complex.
The ritual response is restructure, flatten, cut.
It works on a slide.
Then costs creep back, meetings multiply, and someone books the next restructure. McKinsey found one organization that, after the boxes were redrawn, still duplicated 35 percent of its decisions, held 60 percent more meetings than peers, and spent more than a thousand hours a month on manual reporting.
Structure is visible.
Behavior is not.
AI is the other brass section.
Eighty-eight percent of organizations are experimenting. Eighty-one percent report no meaningful gain on the bottom line. Eighty-six percent of leaders say their organizations are not prepared to put AI into ordinary operations. One in six has no clear owner at the top.
The barriers are not missing software.
They are fear of the tool, legal weather, and the organization's own inability to change how work moves.
McKinsey names three tectonic forces: technology with AI at the center, geopolitical fracture, and a redefinition of work.
The useful phrase buried in the report is from structure to flow.
The twentieth-century firm optimized boxes and lines.
The next advantage is metabolic: how work, context, decisions, and learning actually travel.
Transformation, Geyer writes, is no longer a project you launch. It is a permanent capability. Business as change.
Resilient people and adaptive structures need each other. A rigid organization burns people out. Exhausted people cannot make an organization adaptive.
That interdependence is the part most decks skip, because it will not fit on a pyramid.
The office was never the point
We keep arguing about the office as if the argument were about real estate.
AI does not need a desk.
Routine knowledge work does not need a commute.
What still needs a room is trust, the kind of conflict that improves a decision, apprenticeship, and the sense-making that dies in a forty-five-minute call.
As machines absorb solo throughput, the purpose of a room inverts.
Occupancy stops being the metric.
Collective effectiveness becomes the metric.
The office does not vanish. It stops being the default container of work and becomes an instrument.
You gather when the work is high-bandwidth, high-trust, or high-stakes. The rest of the week is time zone, tools, and memory.
Address on the signature matters less than when we convene, and what we convene for.
This is not remote-work nostalgia, and it is not a return-to-office sermon.
Both were loud songs about furniture.
The undersong is about which human work still requires presence, and which presence we have been wasting on work a machine can already do.
You can already see the replacement in fragments.
A three-person firm where one person carries the relationships, another carries the judgment, and a small constellation of agents carries research, drafting, monitoring, analysis, and memory.
They do not need an IT department, a strategy department, or a floor of coordinators.
They need clear permissions, good instruments, and somewhere worth gathering when the work becomes human enough to require a room.
That is not a miniature version of the old company.
It is a different species.
What wants to replace the firm
Not a slogan.
A different operating system.
Call it a mission mesh, or a collective intelligence, if you need a name.
The successor to the company-as-headquarters looks less like a pyramid and more like this:
Shared memory instead of a shared building.
The filing cabinet, the drive, the "who was in that meeting" problem — all of it collapses into a living graph: artifacts, decisions, constraints, provenance, open questions.
People and agents query the same memory.
Context does not die when someone leaves the channel.
Without persistent memory, you do not have collective intelligence.
You have faster email.
Work charts, not org charts.
Teams form around outcomes and dissolve.
Humans work alongside specialized agents: planners, reviewers, researchers, sentinels.
Management begins to include machines as well as people.
Staffing starts to resemble a film production more than a factory department.
The org chart was always a lagging diagram of last quarter's politics.
Humans above the loop, not in every loop.
The unreadiness of large organizations is partly leaders still inserting people into decisions machines can already make, while under-investing in the decisions only people should make: purpose, risk appetite, harm, capital, relationship, care.
Stewardship is not occupancy of every workflow.
Taking a human out of a routine loop is not the same as taking a human out of responsibility.
Coordination by protocol, not by hallway.
Strategy becomes a living protocol.
Decision rights get written down.
Evidence must travel with an agent's output.
Uncertainty must be disclosed.
Some gates stay human and mandatory, not ceremonial.
Hallways were a lossy protocol that only worked when everyone was already in the building.
And permission becomes architecture.
Collective intelligence cannot mean collective access to everything.
Some context belongs to a person. Some belongs to a team. Some belongs to the institution. Some can safely become signal without ever becoming visible.
The system has to know the difference.
Who contributed something? Who owns it? Who may see it? Who may reuse it? Under what conditions can private knowledge become collective signal?
Without answers to those questions, shared memory becomes surveillance.
The supposedly post-bureaucratic organization simply invents a more efficient panopticon.
Membership more like a network than a headcount.
Contributors assemble for missions.
Some are employees. Some are specialists. Some are agents. Some are community members with skin in the outcome — families, neighbors, trainees, capital partners.
The legal wrapper still exists.
Liability, insurance, regulation, and tax do not evaporate because the work got faster.
The wrapper gets thinner.
The work gets more modular.
Physical nodes, not headquarters.
Studio, clinic, farm, lab, kitchen, workshop, war room.
Places with a job.
Not five-day occupancy as proof of culture.
Leadership, in this shape, stops controlling outputs and starts setting conditions: what gets planted, what the data soil can hold, what gets pruned.
The organization begins to look less like a machine and more like an organism that can sense, decide, act, and write the lesson back into memory before the people who learned it leave.
Why another restructure will not get you there
Restructures fail because they change the slide, not the metabolism.
A collective-intelligence model attacks the actual failure modes in the survey.
Duplicated decisions become objects in a shared memory, with an owner.
Meeting inflation falls when agents draft and humans convene only for conflict and commitment.
Manual reporting becomes a query against instrumentation that already exists.
AI pilots stop being ornaments on an unchanged process and become a reason to redesign the process.
Change stops being a consultant workstream and becomes the default loop:
sense, decide, act, write it back.
Burnout is treated as a design failure, not a wellness webinar.
McKinsey keeps repeating a "double transformation" — technical and organizational.
Piecemeal AI on an unchanged firm is how you get the 81 percent who report no meaningful gains.
Reimagining how work flows is the scarce move.
Buying another copilot and keeping the approval chain is the expensive way to stay unready.
What must not be romanticized
Companies will not disappear.
Offices will not disappear.
DAOs will not save you.
Legal personality still matters.
Someone has to carry liability.
Agents can be copied and reset.
Humans carry consequences.
Any serious model keeps a human residual claimant for harm, not just a prompt for output.
Trust and care do not virtualize cleanly.
Health, aging, education, land, food, and skilled trades still need bodies in places.
A collective-intelligence model that ignores place is a consulting fantasy with good lighting.
The rooms that will matter more, not less, are the ones where someone is taught, fed, housed, examined, or accompanied.
There is also a new bureaucracy waiting: agent sprawl, unowned context, hallucinated institutional memory, indiscriminate surveillance, and "the graph said so" as the successor to "the process requires."
Collective intelligence without governance is a faster rumor mill.
Collective intelligence without permission is extraction.
And readiness is uneven.
A two-person shop with agents and a ten-thousand-person incumbent are not in the same race.
The 72 percent is mostly the incumbents admitting the operating system is wrong and the replacement is not installed.
The refrain
The loud song says organizations are not ready.
The undersong says they are not ready because they are still trying to be companies in the old sense — hierarchies that store intelligence in people, context in hallways, and legitimacy in a building.
What is trying to emerge is not remote work 3.0.
It is a collective intelligence: humans and agents sharing permissioned memory, forming around missions, gathering in places for the work that still requires presence, and treating change as metabolism rather than a launch event.
Most large organizations will try to buy this as software and keep the org chart.
That is how you get another restructure and another thousand hours of reporting.
The ones that pull ahead will treat the graph, the permissions, the protocols, the agent workforce, and the gathering places as one design.
They will measure readiness by whether work flows, not by whether the boxes look cleaner.
So do not start by redesigning the org chart.
Map the work.
Map what has to be remembered, who can know it, where judgment actually enters, which decisions can move without a person, which decisions must never move without one, and where physical presence changes the quality of the outcome.
Then design the organization around that map.
Not the other way around.
The useful questions are not "what replaces the office?" or "is the company dead?"
They are smaller, and harder:
What must be remembered?
Who is allowed to know?
Who is allowed to act?
When must a human convene?
Which rooms are worth the trip?
Answer those five and the rest of the argument becomes a consequence.
The thunder will keep talking.
The drip is the part that tells you the weather has already changed.