Service
Custom AI agents for law firms
Designed with prior risk analysis and strict scoping. No empty promises, no black box, aligned with the Mexican Bar Association guidelines.
Discuss my case →When an AI agent makes sense — and when it does not
An AI agent is not the answer to everything. It is the answer for a repetitive, well-scoped process where the cost of an error is manageable and human supervision can be embedded in the flow. Before building anything, we define the task to automate, the risks the client can tolerate, and what falls outside the agent’s scope. More often than not, the first deliverable of an agent project is not code: it is a scoping document that says "this in, this out, and here is why".
Use cases
What I build them for
Three families where AI agents deliver measurable value inside a law firm.
Lawyer’s personal practice management
Inbox triage, file summaries, contextual calendar, meeting preparation. The agent acts as the first filtering layer before your attention.
Business development and prospecting
Opportunity identification, lead qualification, proposal drafting assistance, structured follow-up. No spam, with human control on every output that reaches the client.
Client communication
Responses to recurring inquiries with escalation to the lawyer, deadline reminders, progress reports. The agent never signs nor takes legal positions — it facilitates the conversation.
Methodology
How I build an AI agent for your firm
Five steps. Each one ends with a verifiable deliverable before moving to the next.
- 01
Discovery — understanding the real work
Interviews with the lawyers who will (or already) do the task. Mapping of the current flow with its exceptions, implicit criteria and friction points. Deliverable: an operational document that any outsider could understand.
- 02
Risk analysis — what can go wrong
Identification of professional risks (confidentiality, privilege, hallucinations, bias), legal risks (GDPR, LFPDPPP, duty of care) and operational risks (dependency, failures, silent errors). For each: probability, impact, mitigation. Deliverable: risk matrix signed by the client.
- 03
Scoping — what the agent does and what it does NOT
Strict perimeter definition: authorised tasks, accessible data, output format, escalation criteria. What stays outside is written down explicitly. Deliverable: functional specification that acts as a technical contract.
- 04
MVP build — minimum viable version
Implementation of the smallest useful version, on real data but in a controlled environment. Auditable systems (no black box), traceability of every decision, human controls at critical points. Deliverable: working agent + full log.
- 05
Operation and improvement — ongoing human supervision
Gradual rollout with monitoring. Periodic review of results with the team. Adjustments based on evidence, not intuition. The agent is never considered "finished": it is maintained like any other legal process, with assigned responsibility.
Commitments
What I never do
- — I do not replace the lawyer’s judgment. The agent is a support layer, never the final decision.
- — I do not use "black box" AI systems whose functioning cannot be audited — aligned with the BMA guidelines.
- — I do not transmit confidential client information to third-party AI services without an explicit no-training contract.
- — I do not bill the client for the time spent learning the tool. Only review, configuration and supervision time.
- — I do not sell abstract efficiency. I sell measurable processes within 90 days with reviewable KPIs.
Frequently asked questions
- What is an AI agent in the legal context?
- A software system based on language models (LLMs) and rules, executing autonomously a sequence of tasks within a scoped perimeter: reading documents, drafting, querying databases, generating structured outputs, escalating to a human when appropriate. It is not a virtual lawyer — it is an operational layer that assists the lawyer.
- Is it safe to use AI agents with confidential client information?
- It depends on the architecture. Public and free services are not safe. It is safe when built on contractually closed environments (enterprise plans with explicit no-training clauses, private deployments, dedicated instances), with anonymisation where applicable and client consent when the law requires it.
- How long does it take to build a legal AI agent?
- A functional MVP on a well-scoped task takes between 4 and 10 weeks, depending on flow complexity, data quality and client team availability. Operation and improvement are continuous — the agent is not shipped and forgotten.
- What happens if the agent makes a mistake?
- Final responsibility always lies with the lawyer in charge, not with the agent. That is why the design includes systematic human control points before any output with legal effect, a traceable log of every agent decision and automatic escalation mechanisms when out-of-scope conditions are detected.
- Do these agents comply with the Mexican Bar Association guidelines?
- Yes — the methodology is designed explicitly to comply. The BMA Guidelines (October 2025) require technical competence, human verification, confidentiality, transparency, non-delegable lawyer responsibility, reasonable fees, and forbid black-box systems. Each of these points translates into a concrete architectural decision, documented in the initial scoping.
Have a process in mind?
The first contact is always a scoping conversation, no commitment. We decide together whether the case justifies an AI agent or whether a simpler solution exists.
Discuss my case →