Deeper cut on the Deloitte Legal and Legora UK strategic partnership. Both sides argue the point is to rewire legal operating models, not just speed up tasks.
Big-client GCs are starting to demand AI usage in RFPs and asking for measurable cost savings. Early sign that the client-side pressure is converting from anecdote to procurement language.
Sobering piece on hallucination rates (69-88% on some legal tasks), professional responsibility, and why even specialised tools need heavy verification. Good risk-side briefing.
Furlong on the widening gap between the interests of law firms and the interests of individual lawyers as AI reshapes the economics of practice. Strong read on where the two start to pull apart.
Phase 2 study pitting 13 AI platforms against in-house lawyers on contract drafting. AI models now match or beat the human baseline on several reliability metrics.
LawNext's write-up of the LegalBenchmarks study. Gemini 2.5 Pro hit 73.3% reliability vs 56.7% for human lawyers, and specialised legal AI flagged compliance risks lawyers missed entirely.
Databricks shows open-source gpt-oss-120b plus automated prompt optimisation (GEPA) beating Claude Sonnet 4 and Opus 4.1 at ~90x lower serving cost. Meaningful signal on the economics of open-model enterprise agents.
Propel tested 45 models on a complex SNAP benefits eligibility question. Latest frontier models now give accurate state-specific answers where older models confidently misled users.
Wordsmith's argument that regulatory and administrative load is now a primary drag on UK growth, and legal AI is part of the answer. Political piece but a useful read for UK-focused conversations.
Argues the comparison market is stuck between reliable-but-limited mechanistic tools and scalable-but-hallucinating GenAI. SuperComparer pitches many-to-many mechanistic comparison as the way out.
Annual pricing report. 50% of firms see rising transparency demands from clients, only 34% have updated pricing to reflect AI efficiencies, and structured matter budgeting delivers ~9% higher realisation.
FT's annual ranking and case studies. Useful view of which firms and projects the FT is crediting this year.
HBR study finding 95% of organisations see no measurable ROI on AI despite heavy adoption. "Workslop" - AI output that looks useful but isn't - is the hidden productivity tax.
Sonnet 4.5 pitched as the best coding model in the world. Significant gains on agentic tasks and computer use, same $3/$15 per million tokens as Sonnet 4.
GDPeval - Evaluating AI on Real-World Tasks (PDF)
[Internal AG resource] OpenAI's GDPval benchmark covering 44 occupations across 9 sectors with realistic deliverables. Close to the R&D benchmarking approach and worth learning from.
How People Use ChatGPT (PDF)
[Internal AG resource] Research on real-world ChatGPT usage patterns - task types, length, sector breakdown. Useful for calibrating internal training and adoption programmes.
ILTA 2025 Technology Survey - Executive Summary (PDF)
[Internal AG resource] ILTA's annual tech survey exec summary. Reliable cross-firm reference for technology spend, adoption and priorities.
$150M Insight-led round plus $260M Accel/Halo-led round. Targeting AI talent and brand with a "Legal Operating Intelligence System" pitch.
SimpleDocs' AI contracting stack plus Law Insider's 5M+ contract, 20M+ clause database across 50+ languages. Big contract-intelligence combination.
Covenant 2.0 unlocks historical legal data for private investors with natural-language querying, portfolio analysis and visualisation. Targets funds, not firms.
TR showcases Westlaw's iterative agentic Deep Research plus a Litigation Document Analyzer, including a feature to spot unverified citations from other parties' work product.
Legal Innovation through Frontier Technology Lab, led by Julian Nyarko and Megan Ma. Combines academic research with firm-vendor collaboration to explore how AI reshapes legal practice and access to justice.