---
title: "Agents change what one unit of work is"
slug: agenti-token-kapital-build-2026
date: 2026-06-15
language: en
status: experimental
canonical_url: /en/2026/agenti-token-kapital-build-2026/
source_slug: agenti-token-kapital-build-2026
translated_from_hash: c6ab9f5b9de49a6d0356748fe7d7fa0e9f8fc27c017f789d8899e5f6c8750a84
translation_status: current
---

# Agents change what one unit of work is

Machine translation of the Czech source.

Main thesis:
- AI work has detached from assistants living at human pace.
- Agents run day and night, have their own teams, and join human teams.
- Human input moves from detailed instruction/specification toward intent.
- AI economics is token-based; companies combine human and token capital.
- Differentiation comes from people, harness, context, and continuous learning.

## Segment 1: Agents change what one unit of work is

Subtitle: Fewer detailed instructions, more context and continuous learning.

1. **Assistant -> Agent** — bigger chunks of work between human interventions. Previously question/quick answer and a small unit of work; now whole code specs, testing, deployment, or Excel -> presentation. More happens between human inputs.
2. **Sync -> Async** — parallel work, even overnight. Humans no longer wait at the screen; they monitor many sessions for who needs help, who has a question, and who is done.
3. **Spec -> Intent** — the human talks to a foreman agent, which prompts other agents. Strong models allow discussing intent; the foreman turns it into specifications and prompts individual agents.
4. **1:n -> m:n** — mixed squads of people and agents. Not just each human with their own agents; mixed teams meet on GitHub for development and Teams for everyday work.

## Segment 2: The new AI economy

Subtitle: Consumption billing and unlocking token value through human capital.

5. **Agents do not sleep** — throughput is no longer tied to human wakefulness. Per-user licensing fit assistants at human pace, like Spotify/Netflix with natural consumption limits; agents are not tied to your time.
6. **Cost vs. value** — humans transform token costs into value. Price depends on model size and vendor; broad model choice reduces risk and opens savings. Value comes from what we can do with tokens.
7. **People and tokens** — the highest gains come from a mix of human and token capital. Either one alone is suboptimal. Gasoline has no value when the car drives in circles; it has value when given direction.
8. **Value is not in the model** — everyone has the same generic model. Value is in unique context, skills, experience, tuning/training, RLE gyms, and human feedback.

Closing:
- AI value will not be in how many tokens you burn, but in how well people can turn them into context, learning, and outcomes.