How can a law firm use AI to track deadlines and case updates without losing control?

AI helps with deadline control when it does the reading: it sorts the court notices and case updates that come in, identifies what each one requires and proposes a deadline with the source passage next to it. The day count stays in a fixed, programmed and tested rule, and no deadline goes into the firm's calendar until a lawyer has checked it.

Have you ever opened the system on a Monday and found dozens of new publications, with no idea which of them start a deadline? This text shows where AI helps in that routine, where it should not decide on its own and how to start with a small scope.

Brazil's Code of Civil Procedure is blunt in article 223: once the deadline has passed, the right to perform or amend the procedural act is extinguished, regardless of any court declaration, unless the party proves just cause. In other words, a missed deadline is rarely fixed afterwards.

The count also has its own rules. For procedural deadlines in days, only business days count (art. 219). The first day is excluded and the last day is included (art. 224). Information is considered published on the first business day after it is made available in the electronic Court Gazette, and the count starts on the first business day after publication (art. 224, §§ 2 and 3). And deadlines are suspended from December 20 to January 20 (art. 220), a period that is coming up soon.

None of this is hard for someone who knows procedure. The problem is volume: many publications, many clients, several people reading and taking notes in different places.

In Franz Kafka's novel The Trial, published in 1925, Josef K. is arrested one morning without knowing what he is accused of. The case moves forward in rooms he never sees, under rules nobody explains, and he spends the book trying to find out where his own case stands.

A law firm does not live a Kafkaesque nightmare, but it knows a smaller version of it: the update arrived, someone read it, noted it in a spreadsheet or an email, and nobody is quite sure whether it became a task. The risk is not in the law; it is in information passing from one place to another.

The DOUVRAS local AI vertical for law and accounting firms has a product for this, LegalFlow, described as case management with AI: deadline tracking, draft petitions and monitoring of case updates. In practice, the work splits like this:

Triage. Separating, in the day's batch, what is only internal court movement from what requires action by the firm.

Reading with source. For each relevant notice, pointing out the case, the client, the required act and the exact passage of text it came from.

Deadline proposal. Suggesting the deadline and due date, calculated by a fixed rule, not by the language model.

Draft for review. When it makes sense, preparing a draft of the filing based on the firm's own templates, for the lawyer to review.

That last point deserves care. A draft is a starting point. The filing that goes into the case belongs to the lawyer.

A language model is good at reading text and understanding what it asks for. For adding business days it is not the right tool: it can get the sum wrong in a perfectly confident tone. That is why, in the design we use, each part does what it is good at:

The model reads. It extracts from the notice the type of act, the date it was made available and the stated deadline, always with the source passage.

The rule counts. A programmed calculation applies business days, the exclusion of the first day, the publication date and the year-end suspension, using the holiday calendar the firm keeps for each court.

The lawyer checks. The proposed date only goes into the calendar after someone has looked at the passage, the count and the calendar used.

This design follows the scope of the DOUVRAS AI agents and assistants service: a knowledge base with declared sources, tests, guardrails, an escalation flow, logs, costs and shutdown criteria. And the handoff between the court system, the email inbox and the calendar is automation and integrations work, with logs, alerts and contingency.

Court notices and case updates carry client names, parties, amounts and strategy. That is the kind of data that calls for care with confidentiality and with Brazil's data protection law, the LGPD. The published guarantees of the vertical are local processing, with open models running on the firm's server and no external APIs; access control by user and by folder; encrypted communication and local backup with the firm's own keys; and usage logs for internal auditing.

Running locally does not by itself fulfill the firm's obligations, as we explain in the text on LGPD, attorney confidentiality and local LLMs. But it removes the most common doubt: where did that document go.

Pick a scope. One practice area or one group of cases, not the whole firm.

Run in parallel. For a few weeks, the AI proposes deadlines while the team keeps tracking them the current way. The two lists are compared.

Test with the past. Use old publications, whose deadlines have already been met, to see where the reading gets it right and where it fails.

Decide when to switch it off. If the proposals diverge from human checking beyond what the firm accepts, the assistant goes back for adjustment or is taken out of use.

Does AI replace the lawyer's deadline control?

No. It reads, organizes and proposes, always with the source passage. Checking and deciding stay with the responsible lawyer.

Does the AI calculate the deadline on its own?

Not in our design. The model extracts the information from the notice, and the count is done by a fixed, tested rule, with each court's holiday calendar.

Can it work with the system the firm already uses?

The project starts with a diagnosis, precisely to map where publications come from, where deadlines are recorded today and what can be integrated.

DOUVRAS serves professional services, such as law firms and consultancies. In a deadlines and case-tracking project, we map the path from publication to calendar, install the model on the firm's server, connect the assistant only to authorized folders and deliver tests, scope limits, logs and shutdown criteria. DOUVRAS is not a law firm: the technical part is ours, the legal judgment is yours. If you want to see the same design applied to documents, read how to review contracts with AI without sending documents outside the firm.

And here is the question: today, in your firm, how many people have to look at a court notice before it becomes a deadline in the calendar?

Douglas Henrique, DOUVRAS Labs.

Brazil. Law No. 13,105 of March 16, 2015 (Code of Civil Procedure), arts. 219, 220, 223 and 224. planalto.gov.br

Brazil. Law No. 13,709 of August 14, 2018 (General Personal Data Protection Law). planalto.gov.br

Kafka, Franz. The Trial. Originally published in 1925.

DOUVRAS Labs. Local AI for law and accounting firms. douvras.com