Early 2025 AI Experienced OS Devs Study (PDF)
[Internal AG resource] METR's RCT with 16 experienced open-source developers. AI tools (Claude, Cursor Pro) slowed them by 19% in real work, even while they believed they were faster by 20%.
The Agent Company - Benchmarking LLM Agents (PDF)
[Internal AG resource] Benchmark of LLM agents on realistic workplace tasks. Top agents only finish 24-30% of tasks autonomously - strong on straightforward work, poor on long-horizon multi-step flows.
LTH roundtable with law librarians on the practical impact of AI on legal research. The people closest to the retrieval problem remain surprisingly calm about where this goes.
Litera launches Lito, an AI agent sitting inside Litera One and intended to act as a virtual team member across both business-of-law and practice-of-law tasks. Big consolidation play across Litera's product suite.
Argues fair AI contract review pricing is well below the $200+/seat mark, with premium pricing reserved for true enterprise features. Useful benchmarking for procurement conversations.
Alex Herrity on where Microsoft Copilot misses nuance in complex meetings, and what "good enough" actually looks like in legal use. Healthy reminder to understand your tool's failure modes as well as its strengths.
"AI is only as good as the data you feed it." Hughes argues firms need to fix data infrastructure before chasing the flashiest AI, or they will underperform those who build the boring foundations.
Harvey and PwC: [Terra Firma Case Study](https://www.harvey.ai/customers/terra-firma) / [PwC M&A Platform](https://www.pwc.com/gx/en/news-room/press-releases/2025/aipowered-deals-pwcs-latest-mna-technology-platform-reimagines-the-way-clients-do-deals.html)
Terra Firma used Harvey's Vault and PwC's UK Tax Model for deal due diligence and tax analysis. Tasks that took days now complete in hours or seconds - a real-deal client example of GenAI embedded in M&A.
Sceptical piece arguing the gap between AI promise and production reality is widening. Useful contrarian read whether or not you agree.
OpenAI rolls ChatGPT Agent into Plus/Pro - it can browse, click, run code and complete tasks on the user's behalf. Signals agents moving from developer API into consumer UX.
Peter Duffy's Q2 2025 round-up: AI adoption lag, strategic M&A and partnerships, Silicon Valley's focus on legal tech, and OpenAI's court-ordered data retention issues.
Marcus's pre-GPT-5 prediction piece: it will be better but still unreliable, still hallucinate, and will not be general-purpose AI. Makes the case that pure scaling has already peaked.
Argues AI agents are now primary users of software and that APIs, schemas, and docs need to be designed for them rather than for humans. Context engineering and curated tools over sprawling menus.
Historical timeline of agentic development in legal tech as a reference point. Helpful for clients trying to work out which vendor claims are real and which are recycled.
Interactive map of 133 legal tech solutions using AI agents, categorised by area. Best single view of how quickly the agent vendor landscape has bloated.
Mollick contrasts the Bitter Lesson (compute beats human priors) with the Garbage Can (real organisations are messy). If the Bitter Lesson wins, companies may not need to map their messy processes before letting AI agents do the work.
European Legal Tech Services Insight (PDF)
[Internal AG resource] European market view on legal tech services - vendors, buyer segments, and adoption dynamics. Good supporting material for European client pitches.
Harvey's self-reported savings: 15-30 minutes per query, 13-25 hours per month for average users, up to 88 hours for top users. Vendor-sourced so treat with care, but useful baseline for conversations on ROI.
Fortune's round-up of the year in legal AI - copilots, research tools, record funding, plus the hallucination, deepfake and bias concerns. Good overview piece for non-specialist readers.
Cohen argues the legal profession needs to shift its default from precedent-driven to forward-looking. Less about the tech, more about the mindset shift required to lead in an AI-enabled market.