> For the complete documentation index, see [llms.txt](https://help.brieftech.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.brieftech.ai/concepts/how-ai-analysis-works.md). # How AI analysis works: agents and playbooks What agents and playbooks are, and what they extract. BriefTech's AI doesn't run as one undifferentiated black box. It is organised into agents, each switched on for a particular matter, and the agents that read and describe your documents are each driven by a playbook that defines what to look for. Understanding that split explains what the AI can and can't do, where its results turn up, and why you stay in control of it. *** ## Agents and playbooks: the two-layer model There are two ideas at work. An agent is the thing that is active for a matter; it acts on that matter's documents — on each one as it finishes processing, or when you run it. A playbook is the instruction set behind an agent that reads your documents — the list of things to extract and how to extract them. Separating the two means the same underlying engine can do very different jobs depending on the playbook it follows. It also means analysis is predictable: an agent extracts what its playbook describes, rather than guessing at the task each time. *** ## The Organise agent and automatic analysis of new uploads The agent you'll meet first is Organise. It reads each document once it is ready and works out what kind of document it is, its key dates, and a short summary. It runs automatically as documents arrive, so a batch you upload and leave alone still comes back classified and described. This is why Review usually has something to correct rather than a blank slate: Organise has already proposed the metadata, and your job is to confirm or adjust it. The **Date** and **Description** fields you see in the Review tab tab are the direct output of this agent — you do not need to trigger extraction manually. Open the Review tab tab after upload to see them filling in as each document finishes processing. The agent also assigns each document a **Document Class**; this is read-only and isn't a Review column, but you can filter by it using the **Classes** group in the left-hand panel. The classes it chooses from, and what each covers, are listed in [Document classes](/reference/document-classes.md). You can re-run Organise on particular documents (for example after correcting a few); re-running replaces only the AI's own suggestions, so any values you've corrected by hand are kept. For the steps, see [Bulk-edit metadata across many documents](/how-to-guides/review/bulk-edit-metadata.md). Your corrections can also steer the AI. If many descriptions are close but not in the style you want, you don't have to fix each one: edit a few by hand, then have BriefTech rewrite the others to match. Your hand-edited descriptions act as the examples and are left as you wrote them. See [Refine descriptions using your own edits as examples](/how-to-guides/review/refine-descriptions-from-examples.md). *** ## What gets extracted — and why the output is structured AI analysis produces structured information — a document class, dates, summaries — rather than loose prose. Structure is the point: because the results land in defined fields, the dates and descriptions flow straight into the Review tab table for checking, and into your Document Index as descriptions and dates. Free text would be harder to trust and impossible to slot into a column. Structured output is what lets the AI's work become part of your finished index rather than a note off to the side. *** ## The AI looks at the document, not only its text When Organise describes and classifies a document, it does not work only from the text pulled out of the file. It also looks at the page itself — the layout, and any figures, stamps, or images on it. That is why a document that is mostly a picture — a photograph, a scanned exhibit, a diagram, or a signature page — still comes back with a sensible description and document class, even when little or no text could be read from it. For example, a photograph — which carries no text to extract — still comes back described by what it shows, in a line such as "Photograph of a damaged car", rather than a blank. You don't switch this on; it is simply how Organise reads a document. It means image-heavy and scanned material gets the same automatic description and classification as a born-digital file. *** ## Why agents are activated per matter Agents are switched on matter by matter, and an agent works only on the documents of a matter it is active in. When you create a matter, BriefTech switches on the agent that reads and describes each document, and the agents behind the **Create Bundle** and **Download Files** buttons on the Index tab, so every new matter starts with automatic analysis already running. Any other agent is activated the first time it is used in a matter — for example, the first time **Duplicate Document Finder** is chosen from the **Tools** menu, BriefTech asks for it to be activated for that matter, and only the matter's owner can do so. Keeping activation per matter means an agent does work only where it has been switched on. The analysis every matter needs runs from the start, while an extra capability such as duplicate detection is available in a matter only once it has been activated there, rather than every capability running everywhere by default. *** ## How analysis feeds Review and the Document Index The results of analysis are not a separate report — they are woven into the pipeline. Extracted dates and descriptions appear in the Review stage, where you correct them, and the values you settle on become the content of your Document Index. Analysis, review, and indexing are three views of the same underlying document details. > For the agents available and what each one produces, see [AI agents](/reference/ai-agents.md). *** ## Related
AI agentsThe AI agents available in BriefTech and what each one produces.AI agents
Bulk-edit metadata across many documentsChange labels, collections, dates, descriptions and index categories on many documents at once.Bulk-edit metadata across many documents
--- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://help.brieftech.ai/concepts/how-ai-analysis-works.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.