McKinsey Technology Trends Outlook 2025 (PDF)
[Internal AG resource] McKinsey's 13 frontier tech trends, led by agentic AI, application-specific silicon, scaled autonomy and deeper human-machine collaboration. Useful backdrop for firm-wide tech strategy discussions.
Thomson Reuters unifies its legal AI under CoCounsel Legal, adds multi-step "Deep Research" with traced reasoning, and embeds agentic guided workflows. Big repositioning of CoCounsel from assistant to agentic platform.
Harvey's long-form view of where legal tech is going: hybrid build-plus-buy, new business models, AI-native competitors, and pricing reform. Worth a read even if you take Harvey's self-interest into account.
Furlong argues lawyers have a commercial role and a civic role - to "bring the law to life" and defend the rule of law. Pointed in the current political environment.
Law.com's chronological walk through Legora's rise, funding and product expansion. Good reference piece to understand the company's trajectory vs Harvey.
International take on whether Harvey's pitch actually lands with UK and EU buyers. Useful counter to the US-heavy funding coverage.
Argues the technology is ready but distribution is the real problem. The channel that will scale legal AI needs to be embedded in the work, trusted by senior buyers, and operationally mature - pointing toward ALSPs.
Building for Production: Generative AI Playbook (PDF)
[Internal AG resource] Playbook on the engineering and operational work required to move GenAI features from demo to production. Useful reference for any internal build effort.
GPT-5 Coverage
Mollick's take: GPT-5 meaningfully changes the UX because the model now picks its own approach, plans next steps, and produces deliverables without precise instructions.
Harvey on how GPT-5 unlocks a "legal coworker" that plans, reasons and executes full workflows rather than one-shot tasks. Vendor framing but useful to see what they're building on top.
Marcus on GPT-5 as an incremental advance. Same hallucinations, same reasoning failures, same generalisation gaps - in his view, vindication of the "scaling is not enough" argument.
Argues AI will eat junior work at the base of the pyramid and force firms into AI-native structures. Bolt-ons won't be enough to hold the partnership model together.
Marveri strips out the chatbot UX and focuses on M&A due diligence - auto-organising data rooms, flagging inconsistencies and missing agreements. Useful counter-example to the "one prompt box for everything" model.
Turning off "training" isn't enough. Any third-party processing of privileged material creates risk, so on-premise systems and informed client consent are the real control points.
Bryter's Michael Grupp argues the legal market has moved from "needs educating" to "urgently buying" and advises founders to ignore the noise and ship fast.
Prompting Science Report 3 (PDF)
[Internal AG resource] Third instalment of the Prompting Science Report covering structured testing of prompt techniques. Useful empirical grounding for any prompt library or training session.
Official guide on prompting GPT-5 - reasoning controls, tool use, agentic workflows, and the quirks of the new router-style model behaviour.
Claude Sonnet gets a 1M-token context window. Matters specifically for long-document legal tasks like full-bundle review, deal data rooms, and multi-volume regulatory filings.
11,000 ChatGPT licences across BBVA with over 80% reporting around three hours saved per week. Concrete enterprise AI rollout data point for financial services client conversations.