A day with LawAgent.

A new case is born already classified: creation covers 28 areas of Brazilian law today, with a primary area and secondary areas that broaden the laws and sources the AI consults — brief structure still follows the primary area. Once created, the user opens the case from Case Management and lands on the Overview — progress, deadlines, and the key information gathered in one place. Alongside it, AI Advisory helps structure the strategy: it has already read the entire case and answers questions like “what are the weak points in our position, in the order the judge will face them”. With the legal ERP connected, this same screen shows the case's complete procedural movement, updated automatically by the firm's document management system.

The Digital Associate is different from the advisory chat: instead of a conversation to answer questions and explore the case, here the user delegates a finished deliverable. Describe the task — an email draft that lands straight in the user's drafts folder, an executive summary of the case, a spreadsheet filled in with the relevant information — and the Associate plans, executes, and returns the finished document for review. Nothing is filed or sent automatically: the deliverable comes back for the user to decide.

With the strategy approved, the brief is drafted in the chosen DNA — that of a specific team member or the firm's own standard. What sets it apart from ordinary text generation is what runs before it ever reaches the lawyer: a chain of AI and deterministic steps that audit one another. Case law is checked against the courts' official API; a precedent citation number that doesn't exist in any verified source is flagged in red as a fabricated citation; a structural validator points out any required sections that are missing; and self-review already fixes what it can, leaving the rest as an open flag. The brief arrives together with this report — showing what was verified, what needs attention, and what still needs to be produced in the case file — instead of arriving as text that merely looks finished.

Strategy and document climb the same approval ladder, configured by the firm: up to four levels, with a named owner at each step and the last one validating and closing out the deliverable. The rule that keeps the process safe is simple — with every new version, approval restarts from the first level; an approval applies to the approved version, not to the document in general. The firm also chooses, per workflow, who produces the work: the team or the Digital Assistant, without changing who approves it. The result is an auditable trail of who wrote it, who reviewed it, who released it, and in what order.

The same case, read from seven different angles — relationship map, metrics, timeline, executive summary, risk matrix, theses and evidence, and an editorial infographic. None of these is a generic illustration: all of them are built from the documents in the case file, with the source of each piece of information indicated, and the user can steer the framing by describing the emphasis they want. They serve both to let the lawyer see the whole structure of the case at once and to bring to a client meeting — and the presentation comes out as a .pptx with one click, with no one building a single slide.

Review covers the other side of the work: the draft that a team member wrote on their own and needs a second pair of eyes before filing. The brief is scored across four axes — admissibility, merits, strategy, and form — and the overall score isn't an average: admissibility caps the result, because a well-built case on the merits is worthless in a brief that won't even be admitted. The AI also locates the opposing brief in the case and builds point-by-point coverage: each allegation in the answer paired with the passage that responds to it and a verdict of sufficient, partial, or unanswered — this is where the omissions that go unnoticed in a quick read show up. Each flag is anchored to the exact passage and comes with ready-to-insert text, and the result comes out as a review .docx or as a redline with tracked changes, to send back to the author.

Drafting contracts starts from an internal library of more than 360 curated clauses, covering a wide range of contract types. The draft is built from that base, with tailored fields and instructions — and whenever the AI uses a clause with no match in the library, it explicitly flags it so the user reviews it carefully. The user can edit existing clauses or add new ones, and with every adjustment LawAgent becomes more aligned with the company's standards and risk appetite.

Still within Contracts, the user uploads drafts produced by third parties — outside counsel or the counterparty — along with the briefing and the strategic guidelines that shaped the work. The AI checks the document against that guideline and returns a complete executive report: an objective recommendation, critical and recommended findings in a table (clause, issue, rationale, and suggestion), and point-by-point adherence to what was requested — separating what was met from what was left out. The report can be shared with the author along with the redline in .docx.

When in-house legal engages outside counsel, the request is born here: briefing, subject matter, deadline, responsible firm, and budgeted fees. The area tracks the status of each request — draft, in progress, delivered — with overdue deadlines highlighted, and receives the deliverable straight into a review. At the top, the financial summary shows total budgeted, total spent, and the percentage of budget consumed, along with the spending ranking by firm. It's useful both for not missing deadlines and for defending the department's budget with numbers. The whole portfolio imports and exports to .xlsx.

Contingencies bring together the portfolio of cases with a prognosis and suggested provisioning under NBC TG 25 — probable, possible, or remote — with total exposure and provisioning calculated over the filtered portfolio. Each case can have its prognosis assessed by the AI, and the quarterly report can be generated automatically at the start of each quarter, with a notice in the alerts. There are three report types: NBC TG 25, the technical-accounting report on provisions and contingent liabilities for closing; Management Summary, the executive view of the portfolio with consolidated exposure and risks; and One Pager, the single-page summary for the board and directors. Each one is exported as XLSX, PDF, or PPT, by period and reference date — and the user can also describe the desired emphasis in free text, knowing the numbers always come from the registered data.

The company describes its own regulatory profile once — activity, markets, sectors, topics — and selects the agencies it wants to monitor among ANEEL, ANVISA, ANATEL, ANP, ANS, ANAC, ANTT, ANTAQ, ANM, ANA, ANCINE, BACEN, CVM, SUSEP, CADE, INMETRO, IBAMA, ANPD, and MAPA. The scan searches for resolutions and ordinances on the official domains of each selected agency and returns alerts already filtered by the profile — not a raw feed of publications. Each alert arrives classified by impact and brings two things that save reading time: the impact analysis, stating exactly what it affects in the company's operations, and the recommended action, with the concrete next step. The user marks it as reviewed or dismisses it, keeping the queue clean.

In an acquisition, the user creates the project, uploads the deal's documents to the data room, and has the AI analyze them. The result is the list of red flags classified by severity and by area — environmental, corporate, contracts, tax, labor — and each finding comes with the exact excerpt from the document that generated it and the read on what it means for the transaction: risk of a shutdown order, early debt maturity triggered by a change of control, a tag-along clause that raises the payout, an under-provisioned liability. It's what turns hundreds of pages of data room into a negotiation agenda and into conditions precedent for closing.

The Legal Help Desk is the entry point for the quick questions that today reach legal by message and eat up the team's time: “can I accept a gift from a supplier worth more than R$ 100?”, “what's the approval authority for service contracts?” The answer is based on the company's own internal policies — not on generic law. It's a single question, not a persistent chat: each inquiry is answered and closed, which keeps the channel focused and frees up the team for the work that truly requires a lawyer.

Once the contract is signed, what remains is complying with what it requires — and that's where deadlines get lost. Obligations extract every commitment from the contract and organize them into two lists: upcoming obligations, with type, owner, date, and recurrence within a defined horizon; and undated triggers, which are obligations conditioned on events — a penalty for delays beyond five business days, a report due within 72 hours of receipt, notice of a change of control 30 days in advance, automatic renewal unless terminated with 90 days' notice. Each item shows the trigger that fires it and the owner, and is completed or dismissed directly from the list.