AI consulting for law firms that keeps the lawyer in charge

A firm may already be using AI without having chosen it. The State Bar of California’s 2026 guidance on generative AI notes that many of the tools lawyers already use, such as legal research platforms, document review systems and practice management tools, include some AI even when nobody labels it that way. AI consulting for law firms starts there: which tools the firm already pays for, where routine work piles up between a first inquiry and a closed file, and which of those tasks a model could draft or sort while the lawyer stays responsible for the result.

What does AI consulting for law firms include?

It follows one path, and each stop on it is its own piece of work with its own page:

• A free worksheet: the AI readiness worksheet lets a firm score itself before it pays anyone.
• An AI readiness assessment: the paid first look at where AI fits in the firm, covered on the AI readiness assessment page.
• An AI roadmap: the assessment’s findings put in order, on the AI strategy and roadmap page.
• An AI implementation plan: the blueprint for one deployment, written before anything is built. See AI implementation.
• AI automation, built and run: where the plan becomes a working system. See AI automation.

A build that needs its own screens and database moves over to custom software, because at that point the work is a new system of record. Our AI consulting overview covers the whole family.

What does California’s 2026 AI guidance ask of a firm?

The State Bar’s Standing Committee on Professional Responsibility and Conduct issued its first practical guidance on generative AI in November 2023. The 2026 version replaces it and, at the California Supreme Court’s request, takes on AI that works with more autonomy: tools that can start tasks, use other software and finish without a person reviewing or approving the result. Three of its points shape AI consulting for law firms from the first interview:

• Responsibility stays with the lawyer: a lawyer remains fully responsible for anything produced with AI’s help, and professional judgment can’t be handed to a tool.
• Supervision grows with reach: the more access and autonomy a system has, the more diligence and supervision the lawyer owes.
• Checks come back around: models change through updates, so competence includes reassessing a system’s capabilities and risks from time to time, and again before it takes on a new legal task.

The limits we would build to are on our law firm automation page. The sections below show where those limits land in a firm’s day-to-day work.

Which routine work could AI take on at a firm?

The four systems we would propose each take over a task someone at the firm does by hand today:

• Intake summaries: the task it replaces is retyping a prospective client’s answers into the matter record and reading back through the emails for every name the conflict check needs. A person still runs the check and decides whether the firm takes the matter.
• Document checklists: it replaces opening each upload to see what it is, then writing the chase email for whatever is still missing.
• Status update drafts: it replaces writing the routine “where does my matter stand” reply from a blank page. The responsible lawyer edits it and sends it.
• Time entry drafts: it replaces rebuilding a day’s time from the calendar and sent mail at the end of the week. The lawyer still records the time actually spent.

Each one drafts or sorts, and a person at the firm checks the result before it leaves the firm or reaches a bill. Which comes first depends on where the hours go, and our guide to what to automate first shows how to rank tasks like these.

What does an AI readiness assessment look at in a law firm?

In AI consulting for law firms, the assessment starts with the work as it is. At a firm, the map we would draw follows a matter from the first inquiry through the conflict check, the engagement, documents, deadlines and billing to the closed file, and marks every place someone types the same details twice.

It should also take stock of the AI already in use. Pearl Technology’s assessment, for one, sets out to find every AI tool in use, “including the shadow AI nobody approved”. At a firm, that list takes in the features inside the research, document review and practice management tools it already pays for, and any tool a lawyer signed up for on their own.

The work differs by practice. Intake at a personal injury firm runs on different questions from intake at an immigration practice, and family law and criminal defense matters each bring their own documents. So in the assessment we would run for a firm, the ranking goes practice by practice: what each step would save, what it would cost to build and run, and how closely a lawyer would need to supervise it.

Before paying for any of this, a firm can get a first read on where it stands from our AI readiness checklist.

Which work stays with the lawyer?

The guidance draws the line at judgment. Lawyers must not set up an AI tool that acts on its own in a way that lets it “make substantive legal determinations, communicate legal advice, prepare and file pleadings, or otherwise act in a representative capacity” without meaningful lawyer supervision and review. Using AI also doesn’t change a lawyer’s duty to make sure the client understands who is giving the legal advice and how the substantive decisions in the matter are made.

In practice, that sets where each system we would build stops. An intake summary stops at a summary, and the lawyer decides what the matter is and whether to take it. A status update stops at a draft, and the lawyer decides what the client hears. A time entry stops at a suggestion, and the lawyer decides what was worked and what gets billed. None of them answers a legal question for anyone.

Research and drafting tools raise the same point from the other side. The guidance says that using AI to research, draft, summarize or generate legal analysis doesn’t diminish the lawyer’s personal responsibility for the truthfulness and legal sufficiency of anything submitted to a court, and that the duty of candor to the tribunal can’t be delegated to AI.

How does AI change what a firm bills?

The guidance answers this one directly. Hourly billing has to reflect the time actually spent on the client’s matter, and AI may shorten some tasks. For more predictable assignments, it suggests a firm may consider flat fees that build the time AI saves into its rates. Whatever the structure, the fee agreement should say clearly how fees and costs are worked out and how far AI factors into the firm’s rates. It sorts what a firm can charge into three kinds:

• Time with the tool: a lawyer may charge for the actual time spent writing and refining prompts, or reviewing and editing what the AI produced.
• General subscriptions: tools for drafting help, research or document review typically count as overhead, like a library or the office computers, and belong inside the fee.
• Costs for one matter: per-use fees for a specialized tool, or a custom AI build for a particular matter, can be billed separately only if the fee agreement says so, at a charge that reasonably reflects the firm’s actual cost, and with no markup or profit unless the client gives informed written consent.

Whether a system we would build counts as overhead or as a matter cost is the firm’s call, made with its own counsel, and the guidance gives examples of both. A time entry draft follows the same rule: it spares the lawyer the reconstruction, and the entry still records the time actually spent.

How should a firm vet an AI tool before staff use it?

Start with what the firm may already have agreed to. The guidance warns that a user may accept a product’s terms of use and privacy policy simply by using it. Depending on the circumstances, it says, reasonable efforts to protect client confidences can include consulting qualified IT professionals or cybersecurity experts, to check a tool’s security, confidentiality and data retention before client information goes in.

Then look at how far the tool can reach. AI tools that carry out steps on their own can be connected to a firm’s email, messaging platforms, document management systems, knowledge bases, client files and calendars. The guidance tells lawyers to carefully evaluate and limit that access, because a poorly configured system may disclose confidential information, including across different matters. So in what we would build, each step reaches only the matter and the folders it needs.

Then keep checking: the reassessment the guidance asks for applies to a tool the firm buys as much as to one built for it. Whether to buy a tool or have one built is a separate question, and our guide to build vs buy software walks through it.

What should a firm’s AI policy cover?

The guidance puts this on the firm’s managers. Managerial and supervisory lawyers should set clear policies on how AI is used, the more autonomous tools included, and lawyer and nonlawyer staff should get periodic training on appropriate use and its safeguards. A subordinate lawyer, it adds, must not use AI tools at a supervisor’s direction in a way that breaks the subordinate’s own professional duties. A policy that holds up names:

• Approved tools: which tools staff may use, for which tasks, and who approved each one.
• Client details: what may go into each tool, and when the client’s informed consent comes first.
• Review: who checks AI output, and at what point in the work.
• Client instructions: how the firm honors a client’s instructions that limit AI on their matters.
• Decisions about people: whether AI plays any part in them. The guidance names the screening of clients or candidates as a use that may carry a risk of unlawful or unethical bias.

An assessment’s map of tools and tasks gives a policy like this something concrete to cover. The policy itself, and any sign-off on compliance, stay with the firm and its counsel.

What would each phase of a roadmap measure?

A roadmap gives each phase a budget and a measure, so the firm can tell whether a step earned its place before it pays for the next one. For the systems we would build at a firm, the measures would be plain ones:

• Time per task: how long intake, document chasing, status replies or time entry take before the system and after it.
• Corrections: how often a lawyer or a member of staff has to fix a draft before it can be used.
• The review queue: how many drafts wait for a person, and for how long.

If a system isn’t saving time once the firm has had a fair run with it, the honest answer is to change it or switch it off.

Where can an assessment lead?

We advise from systems we run ourselves. AlvyOne, the operations system we built with Alvy and still run for them, has AI steps inside it. The models behind those steps get a daily check, and its order intake still changes every week. Alvy is owned by our co-founder Noah Alvarez, and DSI is an Alvy company. That work is freight, and the AlvyOne case study tells its story. For a firm, the same path would end at systems like the four proposed on this page, running inside the tools its staff already use, with its lawyers checking what they produce.

Which AI projects are off the table?

Our AI consulting for law firms has limits of two kinds. Some work sits outside what we build for anyone, and some the guidance rules out for a firm:

• Deciding whom the firm takes on: screening job applicants and other significant decisions about people are outside what we build with AI, and at a firm that covers choosing which prospective clients to take.
• Legal answers for the public: nothing we would build answers a website visitor’s legal question. That advice comes from a lawyer.
• Messages and filings sent on their own: nothing we would build sends a message, a filing or client data out of the firm by itself.
• Model training: we connect models to the systems a firm already uses, and we don’t train models.
• Compliance sign-off: we aren’t lawyers, and we don’t sign off on a firm’s compliance with the bar’s rules. That review belongs to the firm and its counsel.

How does a firm get started?

AI consulting for law firms starts the way every engagement we run does, with a 15-minute call. Our AI consulting cost guide gathers the market bands with their publishers, and a firm’s own figure is set in its written proposal. Midnite Systems was formed in 2023 and works with businesses across the US from our office in Valencia.