The Case for Time Tracking in an Agent-First World

Attention is the new bottleneck in knowledge work

When talking to colleagues in tech and consulting, I hear the same thing over and over again. They are getting more done than ever, yet they feel more exhausted. The to-do list in these industries has always been never-ending, but now we’re not just working longer hours. In the last 2 years, we have switched from working on one task after the other to switching between 10 different agents that need our input anywhere from every 5 seconds to every 5 hours. Our work has shifted from execution to judgement, and judgement done right is much more intellectually challenging.

At Toggl, over 1 million active users track their time, and the data shows this shift:

  1. The share of time entries that mention AI has grown 6.6× since January 2023.
  2. More non-service businesses use Toggl to understand how efficiently their teams work and where the bottlenecks are, not to bill clients.

Time entries mentioning AI

Relative to January 2023

MonthRelative to January 2023
Jan 20231×
Jan 20241.8×
Jan 20252.6×
Jan 20264.2×
Sept 20266.6×

New paying customers

Change in share, 2020 to 2026

Change in share, 2020 to 2026
Not agencies or consultancies+2%
Not agencies, consultancies or software+31%

The role of time tracking has shifted

For decades, the main reason to track time, besides regulation, was to bill clients hourly. That model has a built-in flaw held up only by mutual trust: it incentivizes service providers to take longer and bill more hours than necessary. With AI making various services more accessible, client expectations have changed. Clients are now evaluating whether they should outsource work to service providers or do it themselves with AI agents. As a result, more businesses are moving from hourly to project-based pricing, where finishing faster at the same or higher quality is what drives profit.

This shift from hourly billing to project-based billing changes the reason for time tracking. It is no longer about billing accurately; it is about leverage: the more projects are finished in the same time, the higher the profit. Time tracking now measures efficiency and uncovers bottlenecks. And the biggest lever in knowledge work today is agents. Every business needs to understand where human time goes, which agent opportunities are most promising, and which existing agents contribute most to business outcomes.

The future of time tracking is not tracking hours to bill. It is tracking attention to understand which agents deserve it.

Measuring the attention bottleneck

Working with agents is substantially different from doing the work yourself. Agents are multi-threaded. There is no limit to how many agents you can run in parallel, but our attention remains single-threaded. We can switch between tasks, but we cannot execute in parallel. The more you leverage agents and scale your work with them, the more you will run into this bottleneck, and it can be measured.

Agent idle time describes how much time each agent sits idle after finishing a task, waiting for a human to review or use its output. The higher the agent idle time, the less efficiently your agents run. However, reducing it blindly will result in humans waiting for agents to finish, which is even worse, since human time is still more expensive than agent time.

The goal is to reduce agent idle time as much as possible, not by running fewer agents, but by making agents more autonomous.

Three levels of agent autonomy

  • Level 0 – The agent only runs when you ask it something, like a ChatGPT prompt.
  • Level 1 – The agent runs by itself when a certain condition (trigger) is met, such as when an email hits your inbox. A human still needs to review its output before the agent sends the response.
  • Level 2 – Both the trigger and the output are automated. A customer service agent resolves requests without a human looking at individual runs. Usually, those agents are monitored asynchronously through high-level metrics, like ticket resolution rate in this case.

Moving agents up these levels reduces your agent idle time and increases the number of agents you can run.

Tracking agents

Not every agent belongs in Level 2. An agent resolving simple customer requests is low risk. An agent signing contracts is not. Also, not every agent is worth the effort of making it autonomous. To decide which ones are, you need to understand these three metrics:

  • Agent idle time – How long does each agent wait for a human to come back to it?
  • Human time – Are you dedicating your attention to working with the highest-leverage agents?
  • Outcome contribution – Which agents are contributing most to business goals?

Those who succeed

In the past, those who dedicated their time to the highest-leverage tasks and opportunities succeeded. Today, those who dedicate their attention to the highest-leverage agents will succeed. You need to measure how each agent contributes to your business outcomes. Then you should focus your efforts on resolving the attention bottleneck for those agents.

The future belongs to those who leverage agents best. And they do it by understanding the human attention bottleneck and agent autonomy.