Insights.
Original analysis of legal AI for practising attorneys. Citations to primary sources: cases, rules, and bar opinions linked to bars, courts, and rules. No vendor recommendations. No marketing fluff.
Original analysis on legal AI.
Contract AI and the arbitration shield.
Why transactional AI moved at a different speed than litigation AI for three years. The Federal Arbitration Act section 10 vacatur shield as the structural keystone, JAMS/AAA/SVAMC institutional guidance, and the contract architecture that routes disputes away from courts. A market-structure reading of the $3B asymmetry, and where the dead zone finally closes.
What counts as consent? Reading the 2026 AI cases as a civil-rights question.
The 2026 federal AI cases (Heppner, Warner, Tremblay) are usually read as bar-rules questions. They are bigger than that. The reasoning each opinion uses rests on a theory of consent that, generalised, would unwind boundaries every other consent doctrine in U.S. law has spent two centuries developing. A civil-rights reading.
AI for lawyers: a practical map for 2026.
What "AI for lawyers" actually covers in 2026. The six categories of tooling. What small firms actually use. The five Model Rules duties. The vendor field without the listicle. What stops adoption. What good implementation looks like.
NC State Bar 2024 FEO 1: what it actually requires.
An operational reading of North Carolina's first formal ethics opinion on AI. Which Rules apply. What is permitted. What is required. What the opinion deliberately leaves open, including privilege and tool-selection.
ABA Formal Opinion 512: an implementation playbook.
The ABA's federal-Model-Rules statement on lawyer use of generative AI, mapped to operational practice. Each Rule the opinion interprets, the duty it imposes, and the artefact a firm needs to demonstrate compliance.
Mata v. Avianca, three years on.
A survey of the original Mata sanctions order, the Second Circuit affirmance, and three years of post-Mata caselaw. The verification standard now required of every signing attorney, and the operational practice that satisfies it.
State bar AI opinions: a comparative tracker.
Adopted state-bar opinions and formal guidance on lawyer use of AI: ABA Op. 512, California, Florida, New York, North Carolina, D.C., Mississippi. The common analytical pattern, what multi-state practitioners should track, updated quarterly.
AI vendor diligence: contract clauses to redline.
The contract clauses that recur in legal-AI vendor agreements: training data rights, confidentiality, output ownership, retention, indemnity, breach notification, termination, audit. Per pattern: what to redline. When to walk away.
AI mining judgments for errors: the post-Coney Island map.
AI tools can now read judgments at scale. The Supreme Court's January 2026 decision in Coney Island Auto Parts v. Burton just narrowed what can be done with what they find. The post-Coney Island map: void judgments, fraud on the court, mass-litigation exposure, and the constitutional question the Court left open.
Is ChatGPT confidential for legal work? A 2026 two-layer analysis.
Two layers, not one: vendor confidentiality (what the vendor sees and retains) and litigation discoverability (whether AI use is protected work product or privileged). Tremblay v. OpenAI (2024), United States v. Heppner (Feb. 2026), and Warner v. Gilbarco (Feb. 2026) set the doctrinal frame for both. Same date, opposite results in the two 2026 cases.
Federal Rule of Evidence 707: the proposed AI-evidence rule, read closely.
Public comment closed February 16, 2026; the Evidence Rules Committee voted in May 2026; the rule is on track to take effect December 2027. A close reading of the proposed text, what it changes for trial practice, and the open question of what counts as "machine-generated evidence." Daubert reliability imported into AI outputs.
AI training for lawyers: the actual curriculum a competent practitioner needs in 2026.
Not the marketing version. What the duty of competence under Rule 1.1 actually requires, the five categories of tooling each with its own competence dimension, the doctrinal core of six cases every lawyer should know, the Rule 5.3 firm-governance dimension, and why the single-session AI overview CLE no longer satisfies the obligation in 2026.
AI contract review: a vendor-agnostic buyer's guide for 2026.
Four product categories (drafting, review, full lifecycle, general-purpose with prompts), seven trade-offs, six contract clauses to redline. The highest-spend AI category in legal practice in 2026, evaluated without vendor incentive. Includes the implementation playbook and the wrong-tool-for-the-task discipline.
Best AI for lawyers in 2026: a vendor-neutral evaluation framework.
Every ‘best AI for lawyers’ listicle was written to sell something. The honest framework: six task categories, seven trade-offs, recommendations by firm type, and the minimum-viable stack for a solo at $130-$210/month. Plus why vendor stability is now a real procurement criterion.
AI legal research: CoCounsel, Vincent, Lexis+ AI, Westlaw Edge AI, and Harvey, compared.
The corpus is the product. The verification posture is the differentiator. The pricing is bimodal. The right answer for many firms is two tools, not one. Plus the five-question pilot framework that beats vendor demos every time.
AI legal assistant: what the term actually means in 2026.
The phrase covers four distinct product categories: practice-management embedded AI; general-purpose enterprise AI; legal-vertical research AI; consumer-facing ‘ask a lawyer’ products. Plus the UPL line and an honest read on “free AI legal assistant.”
Legal practice management software 2026: an AI-aware buyer's guide.
The features are convergent. The contractual posture each vendor has signed up to is not. A vendor-agnostic read straight from each vendor’s privacy policy of Clio, MyCase, Smokeball, and PracticePanther, against a six-observation diligence framework: training-data use, sub-processor chains, retention windows, governmental disclosure, anonymisation, and tier differentiation.
Implementation case studies.
Practice-area-specific implementation walk-throughs. The same analysis you'd pay an AI consultancy for, published. Vendor-agnostic; primary sources cited inline; ROI math conservative.
AI in family law: a vendor-agnostic implementation case study.
Family-law practice runs on financial disclosure, parenting plans, custody narratives, and equitable-distribution worksheets. Each is a different AI problem with a different sanctions and confidentiality profile. The full implementation: stack, ROI math, ethics overlay, eight-week playbook.
AI in personal injury: a vendor-agnostic implementation case study.
The clearest single AI use case in any practice area: medical-records review and chronology. The economics flip from paralegal cost-centre to near-zero variable cost, and the contingent-fee margin per case improves materially. Stack, ROI, sanctions-risk overlay, six-week implementation playbook.
AI in immigration law: a vendor-agnostic implementation case study.
The most form-heavy practice in any consumer-facing legal field. AI changes form-assembly and translation economics first; the legal-judgement layer is largely AI-resistant. Detained-client confidentiality and country-of-origin retaliation overlays are practice-specific and matter.
WITH IXSORFIXED-FEE
VENDOR-AGNOSTIC