Today we are giving a name to something Donna has been quietly becoming for a year: an agent. Not a chatbot with a legal vocabulary, but a system that takes a whole piece of work, plans it, does it with the matter’s real documents, checks itself, and hands the result to the professional who signs it.
The word agent is doing a lot of work in the industry right now, and most of it is marketing. So before we show you what shipped, we want to be precise about what we mean, because the difference between a model, a model system, and an agent is the difference between a spellchecker and a colleague.
How we define AI systems at Donna
The building block is a model: a frontier large language model. We use a combination of models from Anthropic and OpenAI, chosen per task. Given a single prompt, a model produces a single response. It is brilliant and stateless: it knows nothing about your matter, remembers nothing, and touches nothing.
Wire a model to task-specific tools and knowledge, and you get a model system. When you ask Donna a question about a contract, a model system retrieves the right passages from the matter’s documents, reads them, and answers with citations. Model systems are one-way trips: a fixed pipeline of model calls and handoffs that runs start to finish, the same way every time.
An agent is different in kind, not just degree. You give it an objective rather than an instruction: prepare the disclosure summary, chase the outstanding conditions, build the settlement timeline. The agent decides what steps that takes, performs them with tools, inspects its own output, and loops back when the work does not hold up. And in Donna, one more thing is part of the definition itself: the loop ends at a person.
Why legal work breaks naive agents
Generic agents demo well and fail quietly. Legal work is uniquely hostile to them for three reasons. The work is grounded: an answer that is not anchored in this matter’s documents is not an answer, it is a liability. The work is accountable: someone with a practising certificate signs it, so “mostly right” is not a grade that exists. And the work is shared: clients, agents and other parties see the output, so the cost of confident nonsense is paid in trust, the one currency a firm cannot mint.
We built Donna Agents around those three constraints rather than despite them.
Grounded: the Space is the context
Every agent run happens inside a Space, the one place where a matter lives. That gives the agent something generic tools never have: the actual state of the work. The contract and its versions, the correspondence, the timeline, the tasks, who the parties are and what has already been agreed. Agents do not search the internet and hope. They retrieve from the matter, and only from the parts of it the person asking is allowed to see. Permissions are enforced at the tool layer, beneath the model, so an agent literally cannot read what its principal cannot.
Verified: cite it or redo it
The verify step in the loop is not a formality. Claims an agent makes about your documents must resolve to citations, and our citations are not footnotes that gesture at a file. Each one resolves to the exact highlighted region of the exact page it came from, so checking the agent’s work takes one click, not one afternoon. Output that fails verification does not get softened language. It goes back around the loop.
Accountable: the loop ends at you
Donna is the hand, not the principal. Anything that changes the state of a matter in a way others can see, sending a document to the other side, filing a task as done on the client’s timeline, committing a redline, stops at an approval gate first. The agent presents what it did, why, and the evidence. You approve, adjust, or reject, and every run leaves a full audit trail: each step, each tool call, each source consulted.
What agents do in Donna today
- Concierge. Ask for an outcome in plain language and Donna works out the steps: “get the building and pest condition sorted” becomes retrieval, a summary of where it stands, the draft chaser, and a task with a due date, presented for your approval.
- Workflow Agents. Build a repeatable process once, in plain language, and run it on every matter: a disclosure review, a settlement checklist, a client onboarding sequence. The workflow carries your firm’s way of doing things; the agent carries the labour.
- Document work. Drafting, comparison and review that runs against the matter’s own documents, with every assertion cited back to its source region.
What is next
The loop gets longer and the gates stay put. We are extending agents to carry multi-day work: watching for the other side’s response, picking the thread back up, and keeping the timeline honest while you sleep. The principle does not move. Donna does the work; the professional owns it.
Donna Agents are rolling out to all plans now. Open a Space and ask for an outcome instead of an answer.