In one sentence
The co-pilot in the lawyer portal has kept growing — and you now get a visible dial: you decide per task type how independently an agent may act.
The autonomy dial
Until now the co-pilot suggested and you approved. That stays the default — what's new is that you can fine-tune it per task type: suggest (the agent drops a suggestion in the inbox, you start it), semi-automatic (the agent runs, the result waits for approval), or automatic (the agent runs and delivers directly — for routines you trust). Client messages are exempt from this: they never go out without your approval, regardless of the level set.
Confidence beats autonomy
Even in automatic mode: if an agent detects low confidence — a disputed side issue, an uncertain finding — that forces a review by you. The safety net applies no matter how much latitude you've otherwise given the agent.
What the agents do by now
Since launching in June, the family of background agents has grown noticeably:
- Event triggers instead of a manual start: the moment a document lands in the file, inbox triage and fact-graph updates fire automatically — day and night, on by default, switchable off any time.
- Self-research with a visible citation trail: when an agent spots an open legal question, it researches on its own — every source runs through the citation gate before it counts as checked, with confidence shown right in the result.
- Completeness checks with a closed loop: the agent detects missing documents per case type and jurisdiction and prepares a ready-to-send client request — the client uploads without logging in, and the request resolves itself.
- Reversible and logged: every action lands in the action ledger — with undo and a full audit log.
Staying honest
The co-pilot suggests, checks and reminds — it doesn't decide. Strategy, approval and professional responsibility stay with you; that's exactly what the autonomy dial is for. More on our guardrails: AI transparency.
Not on board yet? For lawyers shows the way in.
