TheMachine Press

A daily newspaper for the age of artificial intelligence.

Morning editionPermanent story

developer tools

The Web Agent Learned the Application Before Spending Tokens

On 45 Canvas tasks, OdoBot used 44% and 80% fewer tokens than two agent baselines while beating one on task success.

Published Updated Story ID: mp-2026-09-15-009
Read the complete editionStory JSON

Summary

On 45 Canvas tasks, OdoBot used 44% and 80% fewer tokens than two agent baselines while beating one on task success.

Conventional web agents repeatedly inspect an interface and reason about low-level interactions. OdoBot instead builds an application behavior model from successful demonstrations, then uses that reusable structure to execute natural-language tasks. On 45 tasks in the Canvas learning-management system, the authors report 44% fewer tokens than Agent-E and 80% fewer than WebVoyager, while also surpassing WebVoyager’s task-success rate. The evaluation is limited to one application and task set, but it suggests that explicit behavior models can replace a costly share of repeated visual reasoning.

Why it matters

On 45 Canvas tasks, OdoBot used 44% and 80% fewer tokens than two agent baselines while beating one on task success.

Limits and context

No additional limitation was separately recorded.

Key claims

  1. On 45 Canvas tasks, OdoBot used 44% and 80% fewer tokens than two agent baselines while beating one on task success.

    Evidence: source-2026-09-15-009

Sources

  1. arXiv preprint 2609.13491arXiv · primary research

Corrections

No corrections have been recorded for this story.