AI Gameplan · A perspective for discussion
Thoughts on AI for a mid-sized California law firm
My view of where AI could help a firm of about 200 lawyers across California, grounded in fifteen legal tools I have already built and tested. Leadership knows the firm far better than I do; this is meant to start a conversation.
In short
- Start with a clear policy and one approved tool. Most risk comes from people using unapproved tools because there is no good approved one.
- Lawyers decide; software checks. AI drafts and finds. Code verifies quotes, citations and numbers. A person approves everything that leaves the firm.
- Start with tools that already exist. I have built fifteen working legal tools; three (Guideline Guard, Party Finder and Marketing Studio) could be tried right away, with no system integration.
- Match tools to data sensitivity. Client data only in approved enterprise tools; the most sensitive data only on the firm's own hardware.
- Longer term, consider one firm-owned platform that connects every system under the firm's own rules.
Based on public descriptions of a real California firm, which is not named. I don't know the firm's current tools or plans, and some of this may already be under way. Not legal or ethics advice.
1 · What I have already built
Fifteen working tools, each aimed at a real firm problem
All fifteen were built in my own time. Each can be opened and tried in this portfolio; all run on fictional data, and the results below are from recorded test runs.
Ready to try now
These three need no connection to firm systems: each runs in a browser on an approved Claude account and works from a file the firm already has.
- Guideline Guard: a client's billing guidelines plus a prebill, and it flags the entries at risk before the invoice goes out.
- Party Finder: intake documents in, a full list of names to run through conflicts out, each tied to its source.
- Marketing Studio: bios, client alerts and RFP answers drafted in the firm's voice and checked against attorney advertising rules.
Billing: protect revenue before the bill goes out
| Tool | Firm problem | Who it helps | In testing |
|---|---|---|---|
| Guideline Guard | Invoices get cut for breaking a client's billing guidelines. | Billing partners, finance | On a 32-entry prebill: $19.3k of $26k at risk found, $7.8k recoverable. |
| Hour Catch | Lawyers lose billable time they reconstruct from memory. | Every timekeeper | One demo day: 7.3 hours found versus 4.1 remembered. |
Conflicts and intake: faster, safer new matters
| Tool | Firm problem | Who it helps | In testing |
|---|---|---|---|
| Conflion | Letter-for-letter billing-system searches miss aliases, parents and misspellings. | Conflicts staff, intake, GC | Full-function firm system built on conflicts and intake: 3,554 tests; search of 250,000 parties in 5 ms; runs on firm hardware. |
| Party Finder | Intake forms list a few parties; the documents name many more. | Conflicts staff, requesting attorneys | 28 parties found versus 3 on the form (litigation); 33 versus 3 (M&A). |
Litigation: catch errors and find leverage early
| Tool | Firm problem | Who it helps | In testing |
|---|---|---|---|
| Citation Gate | Courts are sanctioning briefs with fabricated or misquoted cases. | Litigators, paralegals | Caught all 4 planted problems in a 7-case test brief. |
| Win your case ASAP | The first read of a new complaint is slow and on a deadline. | Litigation teams | On 26 real federal cases: winning ground in its top three in 14; 357 of 358 quotes exact. |
| Testimony Check | Contradictions between a witness and the record are found late, if at all. | Trial teams | 9 contradictions in about 4 minutes on a 13-document file; all 32 quotes verified. |
Practice groups: the first read of a new matter
| Tool | Firm problem | Who it helps | In testing |
|---|---|---|---|
| Matter Desk | First reads are repetitive and easy to rush. | Real estate, health care, employment, construction, corporate | Lease abstracts, BAA review, PAGA notices, insurance tenders and diligence; all 66 quotes verified. |
| Synthesis | Disclosure schedules and markups are checked by hand under deal pressure. | Corporate and M&A | Demo deal: 7 schedule mismatches, 7 major markup issues, 4 drafting errors. |
Also built: Marketing Studio (bios, alerts and RFPs checked against advertising rules), Praxis (choosing which AI ideas to pilot), LexAtlas (relationship graph with ethical walls), VigilMark (trademark watch), Case Wins Debrief (lessons from wins) and Skippy (a private legal assistant).
For everyday drafting and summarizing, an enterprise edition of Claude, ChatGPT or Microsoft 365 Copilot is worth evaluating alongside these.
2 · Why now
The rules are clear enough to act on, and clients are asking
- Ethics guidance exists. The State Bar and ABA Opinion 512 map AI onto existing duties: competence, confidentiality, supervision, candor and fair fees.1,2
- Courts are sanctioning mistakes. Noland (Cal. Ct. App. 2025): $10,000 for AI-fabricated citations.3,4
- Clients are newly regulated. Employers (FEHA automated-decision rules), health care (AB 3030, SB 1120) and businesses (CCPA rules) all have AI questions, which is also an advisory opportunity.5,6,7
3 · Ground rules
Four rules I would start from
- A lawyer reviews all AI output. Judgment is never handed to a tool.1
- Verify before anything leaves the firm. Every citation, quote and number is checked against its source.
- Bill fairly. Actual time only; no charge for time saved or for learning a tool.1,2
- Measure, then scale. Pilot with a baseline before rolling anything out.
4 · Safeguards
Match each tool to how sensitive the data is
| Data | Where it may go |
|---|---|
| Public | Any approved tool. |
| Firm internal | Enterprise tools that don't train on firm data. |
| Client confidential | Approved enterprise tools with limited retention, single sign-on and audit logs, subject to client instructions. |
| Restricted (health data, walled matters) | Only tools approved for it, or a model on the firm's own hardware. |
Before buying any AI tool, get written answers on: training on firm data, retention, security certification, ethical walls, citations to sources, audit logs, and data return on exit.1
5 · A possible sequence
Four steps, in order
One way to picture it; the pace would depend on what the firm already has and on leadership's priorities.
Step 1
Foundations
Decide who owns AI decisions, publish an interim policy, and survey what people already use.
Step 2
One safe tool
Vet and license an assistant for a pilot group across offices; short training for everyone.
Step 3
A few measured pilots
Test three or four of the tools above on past matters against a baseline; scale, change or stop each.
Step 4
Connect and offer
Connect proven workflows to firm systems, and consider AI governance advice for clients.
6 · A longer-term idea
One firm-owned platform
Eventually, one in-house layer could connect documents, email, conflicts, time, billing and docketing, with the firm's own permissions, walls and audit trail applied everywhere.
- Why: control of client data, answers that draw on every system at once, and freedom to switch AI models.
- What stays the same: existing systems remain the systems of record, and lawyers still decide.
- How: build each quick win on shared parts, then connect systems one at a time, with a go/no-go at each step.
This is a direction to weigh, not a commitment to make now.
7 · Open questions
What I would want to know first
- What AI guidance or tools does the firm already have?
- Which groups feel the most pressure on time or write-offs?
- What are clients asking about AI?
- Which tools does the firm already license that could do more?
- How should early success be judged?
Where I can help
- Run a ready tool on a past matter. For example, Guideline Guard on a closed prebill or Party Finder on a past intake, to see real results before any decision.
- Set up a fair test of any tool the firm is considering, including vendor products.
- Walk anyone through how these tools work, where they fail, and the safeguards they need.
The slides
The slides
Five slides: what I have built, how I use AI, Conflion as a full example, what I could build next, and how I can help the firm with AI.
Sources
- State Bar of California, Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (November 2023).
- American Bar Association, Formal Opinion 512 (July 29, 2024).
- Noland v. Land of the Free, L.P. (Cal. Ct. App., Sept. 12, 2025); summary in McGuireWoods, California Appellate Court Issues $10K Sanctions.
- California Rules of Court, rule 10.430, Generative artificial intelligence use policies.
- Mayer Brown, California Adopts New Employment AI Regulations Effective October 1, 2025.
- Hogan Lovells, New California laws impact uses of AI by health care providers, insurers and vendors.
- California Privacy Protection Agency, California Finalizes Regulations to Strengthen Consumers' Privacy (September 23, 2025).