Presence Is the New Currency

I am writing this in the middle of a contradiction.

I built about twenty AI agents. They research, analyze, write code, review work, and track my projects through the day. In theory, they should give me time back.

In practice, I carry my laptop into elevators. Too often, I keep it beside me at dinner with my family. An agent may finish at any moment and need clarification, feedback, or approval. I tell myself that I am preventing a bottleneck. If I do not respond, I become the blocker.

But if technology saves me time while training me to be reachable every minute, what exactly has it saved?

We measure how quickly models respond, how many tasks agents complete, and how much output teams produce. We rarely measure whether any of that saved time returns to human life.

That is why I believe presence is the new currency.

Presence is more than being physically near someone. It is being mentally available to the person or moment in front of you. It is watching your child perform without checking whether an agent has finished. It is sharing dinner without keeping part of your attention inside a queue of pending decisions.

For technology that promises to save time, presence should be a first-class outcome.

AI can create time. It does not decide who gets to keep it.

Watch the video version of this article.

Saving time is not the same as returning it

AI’s productivity gain is real. I experience it every day. In one randomized workplace study, active users of a generative AI tool spent about two fewer hours per week on email and less time working outside regular hours. But the overall amount and mix of their work did not change. A task became smaller without the job becoming smaller. Shifting Work Patterns with Generative AI

That is an encouraging result. It also reveals the larger question. When a tool saves two hours, who gets them?

The surrounding organization can return that time as focus, recovery, learning, or life outside work. It can also fill the space with more assignments and a higher expectation of availability. Most productivity tools do not make that choice. They accelerate the system they enter.

The same tension appeared in Lenny Rachitsky and Noam Segal’s survey of people working across technology roles. Most said AI was making them better at their jobs, yet more than half reported significant burnout. Their biggest worries were not about direct replacement. They were about rising output expectations and an unsustainable pace. The survey does not prove that AI caused the increase, but it captures the contradiction many of us feel. How tech workers are feeling in 2026

The report calls it “smiling exhaustion.” AI has made building exciting again. It has also made it difficult to feel finished.

When every idea is cheaper to attempt, more ideas become projects. When every project moves faster, yesterday’s exceptional pace becomes today’s baseline. When agents can work at every hour, we are tempted to supervise them at every hour.

The problem is not that AI failed to make us productive. The problem is that productivity became the destination.

The human should not become the agent’s help desk

Many agents work well while the path is clear. Then they encounter an ambiguous requirement, missing information, recoverable error, or choice among reasonable approaches. They stop and ask the human.

Sometimes that is exactly right. An agent should escalate a consequential decision, a request for permission, meaningful risk, or a gap it cannot resolve without inventing facts.

But not every uncertainty deserves an immediate interruption.

An agent should often investigate further, consult a specialized agent, test competing explanations, inspect its own work, recover from a routine mistake, or collect several non-urgent questions into one review.

If it hands every uncertainty back to a person, the visible task may be automated while the coordination burden grows. The human performs fewer steps but remains responsible for watching all of them.

Human in the loop should mean human at the moment of judgment, not human on call for every step.

A long-horizon agent should carry the intermediate work required to reach a responsible decision. Given a goal, it should be able to:

• break the work into meaningful parts;

• use specialized agents to investigate bounded questions;

• compare evidence and test its conclusions;

• recover from ordinary failures;

• batch non-urgent questions; and

• return with findings, options, evidence, and unresolved risks.

The human still defines the goal, permissions, budget, success criteria, and stop conditions. The human still owns consequential decisions.

Long-horizon agents do not remove people from the loop. They move people to the right part of it.

Build AI that can wait

This belief shapes our discovery and analysis work at UserApproved: agents should investigate and synthesize before asking a person to judge. We have intentionally not extended yet that approach through automatic implementation, experimentation, and traffic ramping. That remains the longer-term direction, within explicit human permissions and stop rules.

For any organization building with AI, the practical questions are simple:

• How often does it interrupt a person?

• Which escalations genuinely require human judgment?

• Can non-urgent questions wait and arrive together?

• Where does the time saved by the system actually go?

Measure an agent’s attentional cost alongside its output. Count interruptions and after-hours escalations. Separate questions that required judgment from those the system should have resolved. Track whether people receive longer periods of uninterrupted work and can disconnect.

Do not tell me only how many hours your AI saved. Show me how many of those hours people were allowed to keep.

Presence still requires a choice

Better systems will not make the choice for us. I still have to close the laptop and decide that coaching my children’s basketball team, watching a performance, or playing beach soccer with them matters more than answering a non-urgent question.

Technology cannot decide what I value. But it can respect that decision or continuously compete with it. The future I want is not one where every leader spends the day unblocking one hundred agents. It is one where agents keep working without pulling us away from the people in front of us.

The test is simple: can I close the laptop, trust the system to continue, and be fully present until something truly requires me?