# A3 — Overview

> Stage-based web automation that learns each site with an LLM, then runs Playwright stages deterministically with no model in the loop.

**URL:** https://athree.dev/  
**Canonical markdown:** https://athree.dev/index.html.md

## What it is

A3 (Automated Automation, Again) is a web automation framework built by [UBIO](https://ubio.ai). It uses large language models to *learn* how a website works, write typed Playwright stage scripts once, then *replay* those stages in a deterministic runner.

Compared with agentic browsers that call an LLM on every navigation and click, A3 aims to be about **99% cheaper** and **10× faster** on repeated runs (stated for ~1000 automations with different input criteria).

## How it works

1. **Contract** — You define a workflow: mission, input schema, output schema, site folders of stages.
2. **Learn** — An agent observes the live site, writes / repairs `*.stage.ts` modules, and validates outputs against the schema.
3. **Run** — The runner executes stages as ordinary code (Playwright). No LLM is required in the hot path unless self-heal is triggered.

## Positioning

| Approach | Cost at scale | Reliability |
| --- | --- | --- |
| Hand-written scrapers | Low | Brittle when sites change |
| Always-on LLM browser agents | High | Flexible but slow / expensive |
| **A3 learn-then-run** | Low after learning | Self-healing stages + typed contracts |

## Start here

- Product UI tour: https://athree.dev/product
- Mental model: https://athree.dev/concepts
- Pricing: https://athree.dev/pricing
- Commerce / UCP: https://athree.dev/ucp
- LLM site map: https://athree.dev/llms.txt