This is a Path B operator case-study piece. Where named small-firm cases have been published, they are cited with the source link inline. The legal AI space has more published case density than most professional-services categories in 2026, so this piece leans heavily on real named cases rather than corpus-wide stance.

If you run a 1-to-20 lawyer firm and you want to know what your peers have actually deployed in 2026, the published corpus across Harvey, Spellbook, and GC AI (the latter as a named Anthropic enterprise customer) is now substantial enough to read patterns from. Three patterns are clear: contract drafting and review now ships at small-firm scale, in-house counsel workflows have a dedicated AI tier, and the privileged-content line is still drawn at Enterprise data-handling.

The published 2026 corpus

GC AI is listed on the Anthropic customer page as powering “legal workflows for 1,500 companies, saving lawyers 14 hours a week.” That is a published vendor-side claim, not an independently audited measurement, and the figure is an aggregate across in-house legal teams rather than law-firm practice. For a small law firm reading this, GC AI’s relevance is mostly as a benchmark for the order-of-magnitude time saving the published corpus is willing to claim.

Harvey lists named law-firm customers including Thompson Hine, Fox Rothschild, Lowenstein Sandler, Kelly Drye, Hinshaw & Culbertson LLP, Whiteford, Ogletree Deakins, and Polley Faith LLP. The roster is heavy on mid-size and large firms; the smallest names on the public page (Polley Faith) are still meaningfully larger than a 5-lawyer practice. Harvey itself markets a “mid-size firm solutions” tier, which is the published acknowledgement that the smallest firms are not yet the core market.

Spellbook publishes named small-firm customers including Westaway (USA), KMSC Law LLP, and McInnes Cooper (Canadian mid-size). Spellbook is purpose-built for small-firm contract drafting and review inside Microsoft Word, and its named small-firm references are the closest the published corpus gets to “this is what a 5-lawyer firm runs.” Spellbook reports serving 4,000+ legal teams across 80+ countries and having “accelerated over 10 million contracts.”

That is the corpus. There are smaller players (Lawyaw for document automation, Casetext now part of Thomson Reuters, Vincent AI from vLex), but the three above carry most of the public small-firm narrative in 2026.

What the corpus says about where AI ships

Reading across the three platforms and their published documentation, four workflow categories now show consistent small-firm AI deployment:

1. Contract drafting and redlining. Spellbook’s core surface. The lawyer drafts in Word, AI suggests language, redlines against industry benchmarks, and flags missing standard clauses. Spellbook’s published list of contract types covered (sales, procurement, employment, IP, M&A, real estate, privacy) is a good map of where small-firm AI drafting now ships.

2. Document review at scale. Reviewing 200 contracts for a single non-standard clause used to be week-of-work. The current published platforms (Spellbook, Harvey, vLex) do this in hours with the lawyer reviewing the AI-flagged subset rather than the whole stack.

3. Legal research with citation. Both Vincent AI from vLex and Casetext-now-Thomson-Reuters ship “ask a legal question, get an answer with citations to actual case law.” The small-firm angle here is that the answer quality has now closed enough on Westlaw/LexisNexis searches that solo and small firms can use AI research for first-pass work.

4. In-house counsel inbox triage. GC AI’s surface (per the Anthropic case) is in-house teams handling vendor contracts, NDAs, and policy questions at volume. That is not a law-firm workflow, but it shapes the small-firm conversation because in-house teams that previously would have outsourced this work to external counsel are now handling it internally with AI. Small firms that historically lived on this work see the demand shift.

Where the published corpus draws the line

The corpus is also clear about what AI is not yet doing at small-firm scale.

Privileged content on consumer-tier AI. Every reputable published platform draws this line. Spellbook, Harvey, GC AI, and vLex all run on Enterprise-tier model access (Anthropic Enterprise, OpenAI Enterprise, or equivalent) with zero-data-retention contractual commitments. Pasting a privileged client document into the consumer ChatGPT or Claude tier is the practice the published case studies do not endorse, and the ABA Formal Opinion 512 on lawyer use of generative AI codified this in 2024.

Final advice and judgement. AI assists; the lawyer decides. None of the published case studies of small-firm AI use claim AI is making the final advice call. Harvey’s marketing is careful about this; Spellbook’s is too. The unsexy framing in the corpus is “AI as senior associate, lawyer as partner.”

Court filings without lawyer review. The published corpus is silent on AI-drafted filings being submitted without lawyer-of-record review, which is the post-Mata v Avianca legal-malpractice landscape. The risk does not come from AI being wrong; it comes from the lawyer not catching the wrong.

Multi-jurisdiction privilege workflows. Legal privilege rules vary by jurisdiction (US state-by-state, EU member-state-by-member-state). Cross-border firms in the published corpus run separate AI workflows per jurisdiction, not one unified workflow, because the privilege analysis differs.

The 4-lawyer firm: defensible 2026 stack

For a four-to-six lawyer firm asking “what should we run in 2026,” the published corpus suggests a defensible stack of two to three platforms:

Spellbook for contract drafting and redlining in Word. The published Westaway and KMSC Law cases are the closest-fit references to a 5-lawyer firm in the corpus.
One legal research platform with AI citation, either vLex’s Vincent or Westlaw / LexisNexis with their AI add-ons. The published Vincent case studies focus on small-firm research workflows; the legacy platforms are now competitive on AI features and may already be in your subscription.
Optionally, an Enterprise-tier general AI assistant (Anthropic Enterprise or equivalent) for everything that is not contract-shape or research-shape. The published GC AI case is the in-house benchmark; the small-firm equivalent is using Enterprise Claude or ChatGPT for document review, summarisation, and inbox triage with the zero-data-retention contractual posture intact.

What the corpus says you should not run for privileged work: consumer ChatGPT Plus or Claude Pro, free-tier Gemini, or any AI tool that does not publish an Enterprise tier with an explicit no-training-on-your-data clause. The cost of the Enterprise tier (typically $50-100 per seat per month versus $20 for consumer) is the line item that buys the data-handling posture you need for client work.

What we are deliberately not claiming

We are not claiming any of the named firms in the published corpus produced specific outcomes beyond what their AI vendor publishes. Vendor case studies are vendor-published; they are useful as evidence that the workflow exists at named firms, but the outcome numbers are not independently audited.

We are not claiming the GC AI “14 hours per week” or Spellbook “10 million contracts” figures are typical for a small law firm. They are aggregates across the customer base, and the customer base skews toward larger and more AI-mature operations than a 4-lawyer practice.

We are not recommending any specific platform over another within the corpus. Spellbook fits a contract-heavy practice; Harvey fits a mid-size litigation or transactional practice; vLex fits a research-heavy practice; the Enterprise general assistant fits everyone who handles privileged content outside those workflows.

What changes this read

Cadence on this piece is 60 days because the legal AI space is moving fast and the published corpus expands monthly. The three things that would change the verdict:

A small-firm-specific AI platform reaches scale. Currently Spellbook is the closest fit; if a new platform purpose-built for solo and 1-to-5-lawyer firms reaches a published customer count comparable to Spellbook’s current 4,000+, the recommendation simplifies.
State bar associations issue updated AI-use guidance that materially changes the privileged-content line. The ABA Formal Opinion 512 is the current 2024 baseline; state-level guidance varies and is updating through 2026.
A published small-firm case study claims an outcome that contradicts the corpus. This would be evidence that the platform-published numbers don’t generalise, and would shift the framing from “AI now ships at small-firm scale” to “AI ships at small-firm scale with these caveats.”

We will re-test against the published corpus and the Anthropic customer page on or before 26 Jun 2026.