Ted TschoppIn 2001, I sat in Bob’s Big Boy in Burbank, talking with Erik Davis of WIRED about plans to camp out with Tolkien Online fans before a movie premiere. I told him, “People want to participate in the universe they love.” He recorded the conversation.
Take that same sentance and instead of talking about movies, books, and fandom, frame it into a discussion about the future of work. Put that sentence in a meeting with colleagues and it becomes a surprising idea.
Participation means having something to contribute. It means being able to affect what happens. It means that someone would notice the difference between your presence and your absence. An organization can give a person an account, a title, a queue of tasks, and a full calendar, but it doesn’t always give them meaningful work. It doesn’t give them a “universe” that they love.
But that was in 2001. Today things are very different. Today we are building artificially intelligent machines which take on work and be members of that team. They can handle a larger share of the tasks in the queue.
So, what happens to the person?
I think that question is going to require us to reconsider something much older than artificial intelligence. It asks what we believe work is for, what holds a community together, and how much of our current organization exists to compensate for limitations our technology is learning to remove.
The question becomes urgent when we confuse a person’s place in a process with their place among other people.
In the video, Why Bullsh✲t Jobs are (Finally) Dying, the narrator approaches this from the economics of corporate work. They describe layers of people once needed to move information through large organizations. The argument is that digital communication weakened that need, while cheap capital and organizational incentives helped preserve the layers. And now, AI adds another source of pressure on routine reports and summaries.
The useful question the video leaves me with is whether we still need every agreement, process, and approach we inherited. Its broader claims about useless jobs deserve more scrutiny. A company continuing to operate after layoffs does not establish that everything those people did was unnecessary. Maintenance, accumulated knowledge, and relationships can take time to reveal their absence.
But there is something worth following here. We can remove the technical reason for a particular kind of work and continue organizing people around it for years.
In my 2009 writing on collaboration, I described what happens when knowledge cannot be preserved, found, or checked against its purpose. People compensate. They schedule another meeting, ask someone who knows someone, or carry information from one tool into the next. The informal network becomes the way the formal organization gets anything done.
Those people are often holding the organization together. Their effort is real. So is the possibility that we have made them responsible for a problem we should have solved in a different way.
We gave the office electronic forms(files) and electronic filing cabinets(folders). We also kept a great many people busy moving the contents between them. AI makes it harder to avoid asking why.
I think AI may be the final general-purpose layer of Information Age technology.
That is a large claim, so let me say what I mean by it. We have learned to store information, calculate with it, search it, and move it around the world. We are now learning to delegate more of the work of interpreting it and acting on it. We can describe an intended result and give a system some responsibility for finding a way to produce it.
That ability is incomplete and uneven. A system that handles a well-defined task can still fail when the context changes. But the direction suggests a culminating capability: information becomes something we can ask machines to use toward an outcome, with less of the method supplied by a person at every step.
An earlier version of this argument appears in my white paper on moving beyond Industrial Age utility practices. The practical concern there was preserving context as information moves through an organization. AI brings another possibility into view. The information can increasingly carry work with it.
If that is the direction, much of what comes next may look like operational improvement: better memory, stronger reasoning, lower cost, more reliable execution, clearer interfaces, and wider access. The word “operational” can make this sound smaller than it is. A capability becoming dependable enough for ordinary use can change far more lives than its first impressive demonstration.
I am making a claim about the direction of information technology. Scientific discovery remains open. There may be another transformation beyond the one I can see. This argument does not require us to declare invention finished.
It requires us to notice that we already have enough new capability to question the organization of work. An organizing structure, which was put in place to manage the limitations of the Industrial Revolution, may no longer be optimal. Waiting for a more perfect model is a way of postponing decisions about ourselves.
Consider the phrase: agents swarm; humans click.
It captures a difference in how work gets arranged. An agent system can divide a question into parts, send those parts to other agents, and bring their results together. Meanwhile, a person opens a ticket, copies a number into a spreadsheet, asks for access, attaches a document, and waits for the next person to do much the same thing.
That is a description of a working environment, not a limit on human nature. People have always organized around shared problems. Scientific communities, volunteer projects, and open-source software show how much can happen when contributions can accumulate across many participants. We are quite capable of finding one another and building something together. Yet much of enterprise work places that ability inside a succession of private inboxes and departmental queues.
In Agents Don’t Click, I argued that software changes when agents become its operators. The visible interface becomes less central to routine execution, while authoritative records, permissions, and evidence become more consequential. The next question is what happens to the organization built around those interfaces.
Swarming does not make coordination free. Anthropic’s account of building a system with multiple research agents describes a lead agent assigning parallel investigations and bringing the results together. It also describes duplicated effort, coordination costs, and difficulties when tasks depend heavily on one another. Useful cooperation requires the work to be defined and the results to be checked.
There is a lesson for us in that. We must organize around the actual dependencies of the work. Those dependencies determine the order in which tasks should be performed. In order to do that effectively, we need to understand the relationships between tasks. Best practices here include mapping out task dependencies, identifying critical paths, and continuously reworking the team structure and workflow until you have all those dependencies built into the team and the flow of work.
The distinction can be difficult to see from inside a familiar job. A handoff may contain an essential judgment, an obsolete copying step, or both. You find out by involving the person who does it and examining what changes because of their contribution.
We see this early in IT because code, requests, documentation, tests, and changes already have digital forms. Some outcomes can be checked quickly, and some mistakes can be reversed. That makes parts of our work accessible to agents before work that requires a physical presence or carries consequences that are harder to evaluate.
The same pressure can spread as other kinds of work become accessible. Cheaper execution changes which arrangements are economical. Organizations redesign roles and boundaries in response. The choices they make about ownership, responsibility, and the gains determine what that redesign means for people.
That is a causal path toward much of the economic disruption people fear. It is also a place where human decisions still matter.
The dread makes sense. A person may have spent twenty years becoming the one who can assemble the answer. Their authority, income, friendships, and sense of usefulness may all be connected to that role. Telling them that a machine can assemble the answer in minutes raises questions that a productivity chart cannot answer.
An exponential curve makes those questions harder. We can extend a line on a graph beyond our ability to imagine the institutions that would exist at its far end. We put today’s job descriptions, income arrangements, and professional identities underneath tomorrow’s capability and watch them stop fitting.
Then we look for ourselves in that future.
This is the mirror I think we are standing in front of. It reflects the arrangements we recognize and the things we are afraid of losing. It gives us a picture of our present under pressure. It cannot show us every way people might learn to belong and contribute after those arrangements change.
A prediction can correctly identify a failure in the current system without exhausting the possibilities beyond it.
That does not make every warning mistaken. Losing a livelihood is a real loss. Concentrating power in the hands of those who own the machines is a real political choice. Technical failures and misuse need their own answers. No metaphor makes those problems disappear.
But the end of a familiar arrangement does not tell us that human possibility ends there.
Reaching through the mirror means taking responsibility for arrangements we have not built yet. It means asking who gets a say, who gets access, what contribution earns recognition, and how people can remain secure while their work changes. Hope has to take a form someone can live inside.
That is why I keep coming back to community.
Years ago, in Defending the Commons, I used a garden to distinguish a community from the resource it cares for and the value that resource produces. The people, their shared ground, and its fruit are different things. A useful organization has to care about all three.
In knowledge work, that shared ground includes what we know, how we arrived at it, what we have tried, what failed, and which promises we have made. It includes the experience of the person who notices an exception before it becomes an incident. It includes the conditions that let a newcomer ask a question without first earning entry to an insider network.
We can call this a living commons: knowledge that people can find, question, improve, and use within legitimate boundaries. Someone must maintain it. Someone must notice when a once-correct answer becomes wrong. Its value depends on continued participation.
AI can make that commons easier to search and more useful in doing the work. It can also generate so much plausible material that finding what deserves trust becomes harder. A pile of generated documents is not a community’s memory until people have a way to establish what belongs there and why.
This is the distinction I explored in When Output Becomes Abundant. Production can accelerate while judgment and agreement become more demanding. Removing the need to relay a status update does not remove the need to hear an objection, make a commitment, or take responsibility for someone affected by a decision.
That human work deserves time. It also deserves a better setting than an endless procession of meetings whose purpose nobody can quite explain.
The original Agile Manifesto challenged rigid approaches to software development by emphasizing people, collaboration, and adaptation. Its principles included self-organization, trust, and a sustainable pace. There is a great deal there worth carrying forward.
I want to extend that approach to the community of work itself, beginning with capabilities that now exist. Knowledge can persist beyond a conversation. Contributions can arrive across time and distance. Agents can take on bounded work. The results of one effort can become the starting point for the next.
If we began there, what would we choose to value?
Principles for Work in the Age of AI
We begin with knowledge that can be shared, work that can proceed in parallel, and learning that can outlast any one participant. We organize people and agents around purposes the community can understand and help shape.
We value:
- Shared purpose over inherited boundaries.
- A living commons over isolated experts.
- Coordinated and Collaborative autonomy over serial handoffs.
- Shared capability and gains over isolated productivity.
Roles, expertise, handoffs, and individual achievement remain useful. We judge them by how well they serve these principles.
Shared purpose means the people involved can explain who the work serves and what would make it worthwhile. They can also question it. If participation amounts to helping execute a decision nobody is allowed to examine, we have given the old arrangement a more appealing name. The people who do the work and the people who live with its consequences need a voice in defining success.
A living commons means expertise becomes something others can build upon. A useful decision leaves its reasoning behind. A corrected error improves the shared method. The person who documents a difficult exception receives credit for making everyone more capable. Privacy and confidentiality still apply, but access should not depend on knowing the right person to have a drink with after work.
Coordinated autonomy means giving people and agents room to act within a clear purpose and appropriate limits. Independent work can proceed together. Dependencies stay visible. Someone has authority to accept the result, and someone can interrupt the work when it goes wrong. Accountability is a responsibility we name and support, not something we hope will emerge from the group.
Shared capability and shared gains mean asking what happens to the benefit. A team that teaches a system how its work really operates has contributed more than training data. Its members have supplied judgment, history, and ways of recognizing trouble. They should help decide how the recovered capacity is used, whether that means better service, time to learn, new responsibilities, or a more sustainable working day. When a role does disappear, that conversation must include paid time to retrain, support between roles, and how the affected people share in the benefit.
This is where the manifesto becomes more demanding than a call for efficiency.
If AI makes an organization more productive, leaders need to decide how the collective will benefit. That could mean better pay, greater job security, time to learn, or more control over their work. Employees should help shape those decisions. Promising that AI will eventually make their work more valuable is not enough; people need meaningful influence now.
AI agents may work around the clock, but people need reasonable hours and time away. Faster machines should not create an expectation that employees constantly review results, answer questions, or take on more work. Organizations should define what a completed job looks like and recognize its completion. Finishing sooner with AI should count as success, rather than automatically creating another obligation for more human effort.
You can test these commitments without reorganizing an entire company.
Imagine a recurring customer problem that needs a policy answer, an operational correction, and a clear explanation. In the familiar arrangement, three teams prepare three versions of the response. Each waits for information held by another. Giving everyone an AI assistant might make the documents arrive faster while leaving the customer in the same queue.
Now organize the work around one shared outcome with, the relevant evidence, and the unresolved questions visible to the people authorized to work on it. Let agents investigate independent parts of the problem. Bring people together where the work requires judgment or agreement. Preserve the resolution and its reasoning so the next customer who is stuck in this problem can benefit.
The most revealing part may be what happens to the employee who used to carry the case between those teams. That person may know why the handoffs fail, which information is usually missing, and which exceptions cannot be treated as routine. Give them a role in changing the process. Their knowledge can become a contribution to the whole process instead of a skill the process keeps privately consuming.
Then examine the result. Did the customer receive a better outcome? Did the team spend less time recovering the context of the customer’s predicament? Did errors and rework decline? Could a new colleague understand what needed to be done? Did the participants gain useful time and influence over their work?
If all we can demonstrate is that the system produced more text faster, we have more work to do.
I have argued before that AI adoption requires a change in the operating model. The communal question goes further. It asks what the new model owes to the people whose contributions make it possible. It asks whether a more capable institution can also become a place where people have a more meaningful part to play.
There is no guarantee that we will make that choice. Technology gives us possibilities, and institutions decide which ones become way of everyday life. We can use cheaper information work to preserve the same private authority with fewer people around it. We can also use it to widen access to knowledge, bring more experience into decisions, to spend more time making hard decisions, and make participation less dependent on where someone happens to sit in the building or in the organization.
The machinery alone cannot decide which future is worthwhile.
I come back to that conversation at Bob’s Big Boy, where we were planning to bring Tolkien fans together.
There is a particular pleasure in finding people who care about a story you love. You can ask the question nobody else around you understands. Someone remembers a passage you missed. Someone else sees a character differently. What began as reading becomes a conversation, and the conversation gives people something to build together.
That is the possibility I want us to carry into the next age of work. Machines can help us find the information, prepare the work, and make more things possible. We still have to make room for one another in deciding what those possibilities are for. A person should be able to bring what they know, learn from someone else, and have a hand in what happens next.
The future beyond the mirror is difficult to picture. We can begin with something familiar: people gathering around a shared purpose, each with something to contribute. We can build the next part together.
Fellowship seems like a good word for that.
So I invite you, come and join the conversation; pull up a chair!
