2028: Remember when we didn’t have agents?

It’s a Saturday in 2028, and you’re at the beach.

It took two years of hard work to get here: learning to write specs an agent could actually run with, building the reviews and the guardrails, the harnesses for running agents in parallel, figuring out what to check and what to let go. One by one, you moved your life onto the agentic fleet, personal and professional. The work agents are finishing that server migration that would have eaten this exact weekend two years ago. The personal agents are prepping Monday, including a doctor visit for your eldest and buying groceries for the week.

Everything is delegated. You are, finally, doing nothing..

.. for eleven minutes at a time. You reach into the sunflower-covered beach bag and pick up the phone to see where things are since the last check-in. A lot has happened: the migration cut over two of its three servers, the doctor’s office came back with an opening for Thursday, the grocery order went out, and Monday’s briefing is half-assembled in a doc you haven’t opened yet.

Every time you check, nothing is wrong. The migration is still ahead of schedule. The doctor’s appointment landed on a time that actually works. The groceries came in under budget. Thirteen updates since the last check, all good news, three decisions queued for you to make at a later time — blocking some, but not all, of the work. You clear the decisions, skim the rest, put the phone back in the bag, and immediately start falling behind again while your agents keep working.

Your kid asks you to come into the water. “One second,” you say.

Late that afternoon, walking back to the house, you consider actually disconnecting — you know, pausing the fleet until Monday. Really taking a break. Then you start doing the math on the opportunity cost of 36 hours of paused agents.

You don’t pause it.

The dream of ownerless work

That beach day was supposed to be the dream. It’s the pitch we’ve all heard, and honestly, the pitch many of us have made: the agents do the work, and you lay on the beach doing nothing.

The idea of agents that work while we’re “free” is a delusion. Agents are not these ownerless, fully accountable things that can run around doing our work at no cognitive cost to us.

Agents are owned by us (sorry Fable). Not in the legal sense — in the sense that somewhere, there’s a person where the accountability lands. A person who’s accountable for their work, responsible for what the agent did or did not do, who will be affected by the consequences of the agent’s actions, who is supposed to be supervising — whether they actually do it or not.

Agents don’t eliminate the cognitive cost of work. You go from doing work you’re accountable for to being accountable for work you didn’t do. Accountability never gets delegated.

That’s why laying on the beach while your agents work is nothing like laying on the beach when nothing is running. Same towel, same sun, two completely different heads. In one, the world is holding still, and your mind gets to hold still with it. In the other, while you stare at the waves, emails are going out with your name at the bottom, money is leaving your accounts, code is shipping under your credentials, and a calendar you haven’t looked at is filling itself in. The background processes in your head never shut down — one eye open behind the sunglasses, standing guard over work it can’t even see.

You don’t have to take my word for it. You already live a version of this.

What happened since I last checked?

Think about what your relationship with Slack looks like today. You wake up and check Slack. You leave work and check Slack. You have dinner and check Slack. You’re playing with your kids and.. you get the idea.

I’ve never heard anyone say “I love checking Slack all the time, even when I’m not at work!” Much to the contrary, I often hear “I wish I could stop checking Slack and really disconnect when I’m not at work” — particularly from people spending their time on something meaningful, from simple things like having breakfast with family, through hobbies, to even just going to sleep.

Most of those checks aren’t about work you intend to do. You’re not planning to sit down at the computer and crack on with work for another couple of hours.

So what are they for?

Here’s what checking actually is: a sync. You carry a picture of the world in your head, things have been happening that aren’t in the picture yet, and the itch you feel is the gap between the world as you left it and the world as it is now. Checking is how you catch the picture up.

That’s also why “nothing happened” feels so good. Nothing to catch up on — your picture of the world survived your absence. That, cognitively, is what rest is: permission to stop updating, because the world is holding still.

For most of human history, evenings and weekends worked as rest precisely because the world changed slowly while you were gone. You could leave work on Friday, and on Monday the world you returned to was more or less the world you left.

Then the internet and the phone turned that dial to 11. In tech work especially, the world now moves over an evening, over dinner, over a single meeting you skipped — and the phone in your pocket made every one of those movements visible on demand. That’s the world where checking Slack at breakfast became normal.

Agents turn the dial to 1,000. They act around the clock, so every hour away widens the gap. But they also change something the phone never touched: ownership. The drift is now authored by you. It’s no longer the world changing while you’re away. It’s you changing the world while you’re away, without watching yourself do it. Going to sleep now means waking up behind on your own actions.

Won’t better agents fix this?

The obvious pushback: this is all temporary. Agents are unreliable today, so of course we hover. When they get good enough, we’ll trust them, and the checking will stop.

I don’t think it will — and not because agents won’t get good. They will. But things will continue to happen in ways you didn’t expect. A decision you’d have made differently, an email you wouldn’t have sent, a plan that changed completely since the last time you looked.

Not necessarily failures. Just differences between what you thought would happen and what really happened.

And when the outcomes land under your name, “almost always right” is exactly the level of reliability that keeps you checking: too good to hover over, too imperfect to forget about. You want to be in the loop — not because you expect disaster, but because catching a small deviation early is cheap, and discovering it late is expensive.

If trust were enough to let us disconnect, there’d be no such thing as a manager checking Slack from the beach. Managers with excellent teams — teams they hired, trained, and genuinely trust — still can’t leave the phone in the bag. Not because they expect failure, but because things are moving — and the more things are moving, the more chances something drifts from what they pictured, and the closer they feel they need to watch.

Vigilance scales with the rate of change, not with the error rate. That’s why “better agents” makes this worse, not better: every improvement makes your agents do more, and everything they do is one more thing that can go a way you didn’t picture. Even a flawless run doesn’t buy you rest — it just moves the world further from where you left it.

A bazillion agents, running 24/7

Now scale it.

We’re heading toward a world where running agents is cheap. Not one agent finishing a task — fleets of them, running around the clock, because why wouldn’t they? Compute is bought, the backlog is infinite, and every hour your agents sit idle is an hour of output you chose not to have.

Idle agents are waste to eliminate — or so we’re learning to think.

Notice what that does to rest. Today, rest has a neutral price: when you stop working, your work stops. Slack even goes quiet when your coworkers sleep.

But agents don’t sleep. “Nothing happened” is never an answer. There is always something to catch up on.

And AI removes the neutral price too: once your output no longer requires your presence, resting stops being “not working” and becomes “choosing not to produce.” An afternoon offline used to cost barely anything. Now it has a real price tag, denominated in foregone agent-hours, and you can’t help but do the math.

That’s the world the beach story lives in: constant connection because disconnection means falling behind on your own life, and rest that gets quietly rebranded as missed productivity.

We are setting this world up right now, mostly without noticing, one improvement to agentic harnesses at a time.

Remember when we didn’t have phones?

We’ve run this exact experiment before.. and we have clear results.

Every so often I’ll hear “Remember when we didn’t have phones?” Remember leaving work and being unreachable until morning? Remember weekends where nobody could call you for work at all? Remember going out and just.. being out?

It wasn’t all good. Some real productivity was lost, some coordination was genuinely harder. But there was an ability to rest that today requires actual effort to reconstruct — airplane modes, phone-free vacations, digital minimalism, a whole genre of books about clawing back what used to be the default.

Nobody decided to give that up. There was no meeting. We just took the trade one small convenience at a time — and only saw the price in hindsight, once the old default was gone and “reachable always” had become the baseline someone has to fight to escape.

That’s the thing about these trades: the productivity arrives immediately, and the bill for rest arrives years later, addressed to a person who no longer remembers what rest used to feel like.

In two years, we’ll be saying “remember when we didn’t have agents?” Remember when you closed the laptop and nothing was happening? Remember when the world you left on Friday was the world you found on Monday? Remember when doing nothing meant nothing was being done?

That default is alive right now, and we’re trading it away one overnight run at a time.

Peeking from the future to warn you

While I’m not actually writing from the future, I’m closer to it than most people: between all the agents I’m running in my personal and work lives, I’m on the AI early-access list — bugs and all.

And what I see in myself is exactly what the beach story describes. Both of my mentees, independently, also ran into the same wall recently: the more agentic their workflows got, the harder it became to actually stop. I also see coworkers with serious agent setups struggling to disconnect on their vacations.

The struggle to disconnect is already here. Just unevenly distributed.

I don’t have this solved. Because I can run many agents in parallel, I constantly feel the pull of “I should be running more agents.” There’s always more to be done, and more that can be done.

But I’ve written before about deloads — scheduled weeks where I deliberately train lighter so my body can actually recover. The lesson from training is that rest doesn’t happen by default under an infinite backlog. It has to be scheduled, or it gets eaten.

This year I tried extending the idea to my mind: cognitive deloads, weeks where the plan was to run mentally lighter. I failed twice. With work piling up, and the pull from the agents, the weeks never actually got lighter. Moderation didn’t work.

So this week I’m trying abstinence. No agents at all. Not “I’m not checking the agents” — the agents are off. All week long.

Because if checking is a sync, the only real rest is a world that isn’t drifting: nothing running under my name, nothing to catch up on, no math on what the afternoon is costing. And I’m learning to budget the cost of nothing running the way I budget a deload: not as lost productivity, but as maintenance for long-term results.

The opportunity cost is real, and I pay it on purpose — because the goal was never maximum output this week. It’s high performance for decades. In training, a deload looks like lost gains in the moment, but skipping deloads is how you end up with plateaus or injuries, and those cost far more weeks than deloads ever will. Rest isn’t the enemy of high performance. It’s what makes high performance sustainable.

I’ve been on the other side of the experiment before. For years I didn’t carry a phone, and when someone said “remember when we didn’t have phones?” I’d answer, to their perplexity: “I still don’t.”

So in 2028, when someone at the beach sighs “remember when we didn’t have agents running all the time?” I hope I have an equivalent answer.

But I really don’t right now.