A reference architecture

The AI-Enabled Law Firm

An AI-enabled firm does more than license a legal AI platform. It connects governed data, general AI, legal platforms, practice tools, custom applications, and the people and policies that make them repeatable. Use this map to see what your firm has, what is missing, and what to decide next.

Request an architecture review Explore the full map

From Automated Legal. Developed through work inside law firms, including analysis of thousands of lawyer prompts. Updated August 24, 2026 · Version 1.3.

Layer 0Data FoundationGoverned, reachable firm data

Layer 1General LLMEveryday AI beyond legal tasks

Layer 2Legal AI PlatformThe firm-wide legal workhorse

Layer 3Practice ToolsSpecialty tools by group

Layer 4Custom AppsWorkflows built for the firm

Layer 5EnablementMeasurement, skill, governance

A defined engagement

What an architecture review includes

A fixed-scope, paid assessment for firms deciding what to build, buy, integrate, or govern next. Scope and fees are agreed before any firm data is shared.

Current-state map: your stack, owners, integrations, and active pilots across all six layers.
Decision analysis: the gaps, overlaps, security constraints, and dependencies affecting the decision in front of you.
Recommended sequence: a prioritized set of next moves, with the decisions your leadership team still needs to make.

The outcomes · Layer 4

Custom Apps & Agents

Ask a whole case record a question and get a cited answer. Review an NDA in minutes. Pressure-test a brief before you file it. This is the work the rest of the stack exists to produce.

A purchased tool solves the vendor's version of a problem. These are built against a firm's own documents and its own complaints. Some can be prototyped quickly once the layers below are in place. Production use still requires testing, permissions, ownership, and maintenance.

Swipe through the applications

Client Knowledge Agents

An agent on a matter's documents, so anyone on the team can query the whole record and get cited answers.

NDA Generator / Reviewer

Reviews incoming NDAs against the firm's playbook, identifies potential deviations for lawyer review, and drafts a response.

Case Opening Automation

Turns intake notes into a stronger starting draft of the case assessment memo, demand letter, and litigation hold notice for lawyer review.

Judicial-Pattern Analysis / Argue-the-Other-Side

Pressure-tests a draft against documented judicial patterns or the other side's best argument before filing.

Adversarial Argument Simulation / Deposition Simulator

Simulated questioning for witness preparation and associate training, with lawyer oversight.

Document Formatting Agent

Applies a partner's or a court's formatting specification to a draft automatically.

See 12 more applications

Partner Writing-Style Profiles

Codifies each partner's structure and phrasing so first drafts arrive in their voice instead of the model's.

Timekeeping / Billing Narrative Agent

Reconstructs the day from calendar, email, and document activity, and drafts narratives that comply with client billing guidelines.

Matter Deadline / Email Agent

Watches email, calendar, and deadlines across every active matter and surfaces what is due, unanswered, or quietly drifting.

Firm Brief Bank / Precedent Vault

Makes prior work product searchable behind a confidentiality screen, so drafting starts from the firm's best previous version. Usually the cheapest high-value build on this list.

Contract Provision Extractor

Pulls the provisions that matter (venue, arbitration, fee-shifting) from any incoming contract into one scannable view.

Distribution Waterfall / Cap Table Modeler

Extracts economic terms from operative documents and models the distributions and ownership.

Experience Database / RFP Responder

Makes firm experience searchable by industry, deal type, and outcome, and drafts pitches with real matter citations.

Interview / Recording Intelligence

Mines recordings and transcripts for admissions, inconsistencies, and timeline conflicts, cited to the source.

Prompt Library by Litigation Phase

The firm's playbook organized the way lawyers think: by phase of the case, not by feature of the software.

Client Intake Qualifier

Screens inbound matters against the firm's intake criteria before an attorney spends time on them.

Competitive / Client Intelligence Agent

Monitors clients, prospects, and their markets, and delivers the developments worth a partner's call.

Security Questionnaire Responder

Drafts responses to 200-item client security audits from the firm's own policies and control documentation.

One example

How one knowledge agent uses all six layers

Consider a knowledge agent that answers questions across a large matter file and cites the source. The visible application is Layer 4, but it only works when every layer underneath it has a defined role:

Documents worth pointing atLayer 0
a governed, retrievable matter file with retention and access rules applied.
A general model for the supporting workLayer 1
intake summaries, formatting, and the client presentation.
A legal platform to host itLayer 2
the approved vault and agent tooling where the workflow runs.
A practice-group tool where one existsLayer 3
the eDiscovery platform used to prepare the record.
The application itselfLayer 4
an agent configured on the matter's documents and tested by lawyers who can identify unsupported answers.
People and policies that keep it reliableLayer 5
prompt skill, usage measurement, governance, ownership, and a review process.

Every app above needs the same stack underneath it. That is why this is an architecture and not a shopping list.

Layer 5

Operations & Enablement

What separates the firms getting the outcomes above from the firms just paying for licenses.

Licenses are not usage, usage is not skill, and nothing is governed by default. Layer 5 is not a step in the sequence; it is the discipline that makes the other five layers produce. In LexisNexis's 2025 research, half of lawyers said their organization did not measure AI against clear success metrics. That leaves too many renewal decisions resting on anecdotes rather than evidence.

Usage analytics across all your LLMs
who uses which tool, how often, and how well, in one place instead of four vendor dashboards. Measurement comes first.
Personalized prompt coaching
per-lawyer reports built from each lawyer's own prompts. Training rooms do not change behavior; a private report about your own work does.
Prompt libraries by practice group
the prompts that work, written down and maintained.
AI champions and peer-led training
a respected lawyer demonstrating a real use case on a real matter, the only format that survives a busy calendar.
AI governance
acceptable use, disclosure, privilege guardrails, client consent where outside counsel guidelines require it, and a data lifecycle for every tool.
Vendor terms review
retention windows, human-review carve-outs, and amendment rights, compared before signature.
Pilot design and vendor evaluation
head-to-heads with written criteria and a decision date, not pilots that quietly run forever.

Start with measurement, because it is the only step that tells you which of the others your firm needs. Automated Legal turns your firm's own AI usage data into a monthly picture: who is using what, how well, and what changed. When you want help building what measurement exposes, that hands-on work is here.

Layers 0–2

The stack underneath

Why the same outcomes succeed at one firm and stall at another usually comes down to these three layers.

Layer 0 · Data Foundationcloud DMS (iManage Cloud, NetDocuments), cloud email, time and billing systems (Aderant, Elite 3E), retention, classification, DLP, matter data at intake
the layer nobody wants to fund, and the ceiling on everything above it. Much of the disappointment attributed to legal AI is a Layer 0 problem wearing a Layer 2 invoice.
Layer 1 · General LLMClaude, ChatGPT Enterprise, Microsoft Copilot
the everyday AI for everything legal AI does not cover: spreadsheets, email, formatting, knowledge search, staff productivity. Copilot removes adoption friction inside M365; the spreadsheet-heavy lawyers we work with rate the frontier models stronger. Pick with a head-to-head pilot, not a renewal default.
Layer 2 · Legal AI PlatformHarvey; CoCounsel, Lexis+ Protégé, and Legora compete for the seat
the firm-wide legal workhorse: research, drafting, vaults, workflows. The unlock is DMS integration, which turns the firm's own work product into something a lawyer can query in plain language. Know the gaps and re-check them quarterly: redlining, footnotes, spreadsheet work, and cross-session memory have improved on paper, and where they still run thin in practice, that is the work Layers 1 and 4 absorb.

Layer 3

Tool catalog by practice group

The tools only one practice group needs, and roughly what each costs.

Layer 2 covers the firm; this layer covers the practice group, and standardization stops being the goal, because a litigation team and a trusts and estates team have almost nothing in common at this layer. The depth is uneven on purpose: litigation and transactional tooling is years ahead of the rest, and the groups you do not see here, employment among them, are served by Layers 2 and 4 today rather than by a weaker vendor. Vendor names are examples of a category, not endorsements, and this is the fastest-moving layer in the stack: treat the category as durable and the vendor as a snapshot. Scroll sideways to read the full table.

Practice group	Tool (example vendors)	What it does	Cost signal
Litigation	eDiscoveryRelativity aiR, Everlaw; Logikcull at the budget end	AI-assisted review, culling, and privilege screening across large productions	$$$
Litigation	Litigation analytics and judge dataLex Machina, Pre/Dicta	Outcomes, timing, and motion tendencies by judge, court, and opposing firm	$$
Litigation	Chronology and fact managementCaseFleet; CaseMap+ AI if the firm is already on Lexis	A cited fact chronology that keeps the case story linked to the evidence	$
Corporate/M&A	In-Word contract draftingSpellbook, DraftWise	Drafting and clause suggestions inside Word, checked against firm precedent	$$
Corporate/M&A	Due-diligence reviewKira, Luminance	Extracts and classifies provisions across a data room, flagging deviations	$$$
Corporate/M&A	Deal execution and document mechanicsSimplyAgree; Litera, Draftable	Closing checklists and signature pages, plus reliable redline comparison where the legal platform falls short	$
Real Estate	Title, survey, and lease diligenceOrbital	Reads title commitments, surveys, and leases; extracts obligations onto GIS	$$
Real Estate	Lease review at portfolio scaleKira, Luminance	Extracts terms, obligations, and deviations across lease portfolios	$$$
Trusts and Estates	Document assemblyKnackly; replacing legacy HotDocs	Generates estate planning document sets from structured intake	$$
Trusts and Estates	Client intake and formsGavel	Turns firm templates into client-facing intake that generates the document set	$
Tax	Tax research AIBloomberg Tax AI, Thomson Reuters Checkpoint AI	Natural-language research over primary tax authority, with citations	$$
Tax	Tax position analysisBlue J	Models how a tax authority or court would likely treat a position, with the authorities behind the answer	$$
IP	Patent search and landscapingPatSnap	Prior art search, portfolio analysis, and competitive mapping	$$$
IP	AI patent draftingSolve Intelligence, PatentPal	Drafts claims, specifications, and figures for attorney revision	$$
Funds	Private-markets contract automationOntra	High-volume NDAs, side letters, and MFN elections for fund counsel	$$
Funds	Investor onboarding and subscription documentsPassthrough	Electronic sub docs with conditional logic, built with fund formation counsel	$$

Costs are relative to a 50 to 300 lawyer firm: $ a practice group can absorb, $$ a budgeted firm decision, $$$ a committee decision, often matter-billed.

Where to start

Measure first. A metadata export can show adoption by role and practice group, unused seats, feature usage, and movement over time. Prompt analysis is a separate, explicitly scoped assessment.

Pilot your general AI and your legal AI platform (Layers 1 and 2); do not default. Same tasks to each tool, criteria written in advance, a decision date on the calendar. Read the enterprise terms during the pilot, not during the renewal.

Build one Layer 4 app that answers a real complaint. The NDAs, the formatting, the brief nobody can find. One working app does more for adoption than any rollout email.

Already using legal AI?

Request a complimentary Adoption Snapshot to see active users, frequency, feature usage, cohorts, and movement.

Request an Adoption Snapshot

Still deciding what to build or buy?

Request the fixed-scope, paid architecture review defined above.

Request an architecture review

Do not email usage exports or prompt data. We agree the minimum data needed and provide secure transfer instructions after scoping.