Glossary · 28 Sep 2026

What is an AI agent?

← Glossary

An AI agent is an AI system that carries out tasks as well as answering questions. Given a goal, it plans steps, uses tools such as databases, spreadsheets, email or maps, checks results and continues until it finishes or gets stuck. Most current agents are built on large language models. The researchers behind one benchmark describe them as AI agents "that interact with and affect change in their surrounding environments".

How well they work

It depends heavily on the task. TheAgentCompany, a benchmark that simulates a small software company, found in 2025 that "the most competitive agent can complete 30% of tasks autonomously". Its authors conclude that simpler tasks could often be automated, but "more difficult long-horizon tasks are still beyond the reach of current systems".

Domain benchmarks tell a similar story. On 349 realistic GIS tasks with exact answers, a 2026 benchmark found the best agent completed 32.7% under strict scoring, "although most models produce outputs that are close to the ground truth". Both benchmarks use hard, varied tasks. A narrow, well-specified task in one business may go much better, and newer models score higher over time.

In a timber business

Agents suit tasks whose results a person can check: gathering the week's tickets for a contract, drafting a reconciliation for review, chasing missing documents. They suit less well tasks where one wrong step is costly and hard to spot, such as changing prices or approving payments.

What it isn't

An agent isn't a chatbot, which only replies. Nor is it an automation script, which follows fixed steps. It takes actions like a script but chooses them itself, which makes it harder to predict.

What to check

Ask what tools an agent can use and what it can change. Ask for its completion rate on tasks like yours, measured on real examples. And keep a person responsible for anything it does that moves money or commits the business, with a record of every step it took.

How Quarri works explains the platform as a layer over existing systems, not a migration.

Sources

  1. Xu, Song, Li and others, "TheAgentCompany: Benchmarking LLM Agents on Consequential Real World Tasks", arXiv 2412.14161, revised 10 September 2025: arxiv.org
  2. Pothuri, Jiang, Xu and Yang, "GISAgentBench", arXiv 2608.01645, 3 August 2026: arxiv.org

Quarri is an AI-native data platform for the timber supply chain. It connects buying, production, sales and inventory for forest management, sawmill, wood products and pulp, paper and packaging operators.

See it on your own data.

Live in two weeks, on the systems you already run.