News You Can Use

Edition 25 · 1st - 14th Sep 2025

News You Can Use

Deep Dives

Three stories worth sitting with

Artificial Lawyer: What Is MCP and Why You Need It

Artificial Lawyer: What Is MCP and Why You Need It

What
Model Context Protocol (MCP) is an emerging open standard that lets AI systems securely connect with external tools and data. Instead of being confined to text-only responses, MCP allows an AI to perform real actions in a controlled way. The key idea is giving users fine-grained control over what the AI can and cannot do.
So what
MCP is becoming a foundational technology in AI workflows. We are exploring how MCP could help AGPT (and other internal tools) connect securely and consistently with our systems, while also building in safeguards and permission controls.

CCBJ: Why MCP may be the legal industry's next breakthrough

CCBJ: Why MCP may be the legal industry's next breakthrough

What
Paul Walker, Global Head of Product at iManage, highlights that MCP addresses the limitations of traditional, brittle, and insecure integration methods by enabling secure, universal, and agile connections between AI tools and enterprise systems.
So what
MCP offers the flexibility to choose from a range of AI tools without being locked into a single vendor. This interoperability enables firms to customise their AI solutions to meet specific needs.

Thinking Machines: Defeating nondeterminism in LLM inference

Thinking Machines: Defeating nondeterminism in LLM inference

What
The blog post explains why LLMs sometimes give different answers for the same question. The main causes are small differences in how computers handle numbers and the order in which tasks are run. The blog post suggests using "batch-invariant kernels" to make the results consistent.
So what
This means LLMs can be made more reliable and predictable. Guaranteeing consistent results is important for testing and deployment - we are seeing this challenge with some work on AI Benchmarking and reviews through Legora.

Worth Reading

Everything else worth a click

LexisNexis - The AI Culture Clash

UK survey showing 61% of lawyers now use AI but only 17% say it's truly integrated into firm strategy. Cultural resistance and unclear ROI are the main drags.

NYT Opinion - AI, GPT-5, Rethinking

NYT op-ed on what the muted GPT-5 reception reveals about the wider AI narrative. Useful piece for leadership trying to gauge whether the scaling era is pausing.

Hg - SLG 2025: AI in the Real World

Morgan Stanley and A&O Shearman on multi-year AI transformation: infrastructure, domain expertise, human-in-the-loop and a willingness to rewire business models. A&O Shearman specifically called out the shift from hourly to fixed-fee.

Dazza Greenwood - Beyond AI Benchmarks

Argues generic benchmarks don't capture what "good" looks like for your organisation. Firms need custom evaluation frameworks built from golden datasets and domain expert input.

OpenAI - Why Language Models Hallucinate

OpenAI's paper on the training-objective roots of hallucination. Models are rewarded for confident guesses rather than saying "I don't know", which is structural rather than a bug to patch.

Legora x Deloitte Legal - Strategic Relationship

Deloitte Legal partners with Legora for AI-led legal transformation programmes, with custom workflows and pre-built solutions on the Legora platform. Big Four legal consulting wrapping itself around a legal AI vendor.

why-language-models-hallucinate.pdf

[Internal AG resource] PDF of the OpenAI hallucination paper. Useful background reading.

PwC - UK Legal Services Market Report 2025 (PDF)

[Internal AG resource] PwC's annual view of the UK legal services market - revenue, profit, headcount, tech spend. Solid benchmarking data for firm-level conversations.

Working with AI - Measuring Impact of AI (PDF)

[Internal AG resource] Framework for measuring AI impact in a legal business. Useful for setting up internal KPIs beyond "seat count."

The Contract Intelligence Index Report (PDF)

[Internal AG resource] Report on the state of contract intelligence tools and adoption. Useful reference for contract AI procurement and benchmarking.

Medium - How Harvey Built Trust in Legal AI

Case study breaking down Harvey's trust stack (custom legal models, 0.2% hallucination rate, case law verification), its "Minimum Viable Trust" GTM with flagship firms, and the LexisNexis / Microsoft ecosystem plays.

Nature - News and Views on AI

Nature commentary on recent AI research developments. Useful signal for what's landing in the scientific community vs the hype cycle.

Gary Marcus - OpenAI's Future, Foretold?

Marcus argues that without a further step-change in demand, OpenAI's route to $200bn revenue by 2030 looks shaky. Weak ROI stories and underwhelming releases make the compute bill hard to justify.

Juro x Wordsmith - Strategic Partnership

Juro (CLM) and Wordsmith (agent workspace) integrate via MCP. Clients of both platforms get cross-functionality - early example of how MCP can stitch vendor ecosystems together.