Nathaniel Whittemore calls the thing underneath this the infinite backlog, and the logic is simple and uncomfortable. You used to stop working because you ran out of road. You personally could not do any more, so the day ended.
Agents changed that. An agent does not tire, does not lose focus, does not need to go home. The only reason an agent is not working is that you have not given it something to do. So the natural end of the working day quietly disappeared. There was always one more thing you could set running.
We beat the end boss of time. The prize was not leisure
It was a queue that never empties. For the first time, the limit on how much can be done is not your capacity. It is your willingness to keep feeding the machine. And it turns out the amount of work to be done was never really bounded. There was always something next. Now there is always something next that you can actually do.
You can feel this in any team that has gone deep on agents. The work does not arrive in a neat pile any more. It arrives as a standing invitation. Another report you could generate. Another process you could automate. Another backlog you could clear.
The old skill was getting through the work. The new skill is deciding which work deserves to exist at all.
Two traps, sitting next to each other
Here is where service leaders need to be careful, because the instinct is to celebrate all this. Look at the throughput. Look how much more the team is shipping. But volume was never the point, and an organisation that mistakes activity for progress will simply scale its own busyness. We have spent years building cultures that reward output. We are about to find out what happens when output becomes close to free and the people producing it are not.
The second trap sits right next to the first. At the same moment we removed the limit on how much we can produce, we started teaching agents how we work. We feed them our processes, our playbooks, our accumulated way of doing things, and we ask them to scale it.
Hearing everyone is healthy. Averaging everyone is not. If the way we work is the product of relentless consensus, of every idea blended into a paste that nobody objects to, then we are not automating excellence. Dan Shipper calls the default output of this world slop. Visible sameness, repeated endlessly. The infinite backlog filled not with brilliance, but with our own settled average, produced faster than we have ever managed before.
Put the two together and you get the genuinely bad outcome. Exhausted people producing forgettable work, at scale, and someone calling it transformation. Nobody designed that future. It is the one you drift into if you treat AI as a way to do more of the same, only faster.
What is actually scarce now
The way out is to be honest about it. It is not execution. Agents have made doing close to infinite. What is scarce is judgement: deciding what is worth doing, and deciding what counts as good once it is done. Those are the two jobs that do not scale and cannot be handed over. They are also the two jobs that consensus quietly destroys, because a committee fills the queue by reflex and decides by averaging.
So the discipline that matters now is not speed. It is subtraction. The leaders who get real value out of this will be the ones who get good at deciding what not to do, who protect the few things worth doing properly from the infinite list of things they now technically can, and who refuse to let the machine's appetite set the agenda.
Saying no becomes a senior skill. So does deciding, deliberately, what your people should never have to touch.
This is the same fault line I trace in most AI strategies being a detailed plan to become average, faster. There, the sameness comes from the models. Here, it comes from the appetite. Both arrive at forgettable work, and the Human Operating Model exists because the only defence against either is human judgement, deployed deliberately.
For a generation, the question in service operations was how fast can we go. The better question now is where should we point this, and what should we have the discipline to leave undone.
An infinite backlog is not a gift. It is a test of whether you actually know what matters. Fill it with consensus and speed and you get tired teams and forgettable work. Fill it with judgement and you get the one thing AI cannot hand you: work that was worth doing in the first place.