Browse by format: research briefings or engineering guides.

Latest research

How Much Does AI Improve Software Development Productivity?

Strong evidence supports 1.1–1.3× release output; highest-automation cases report a low-confidence 4.5× median and 10×+ upper end.

Read more

Building the Ed-o-meter: Notes on Writing My Own LLM Benchmark

The harness, the tasks, the decisions that turned out to matter, and the mistakes — written down mostly because the mistakes were more informative than the results.

Read more

Locked-Down Fable Disappoints, Cheap-as-Chips GLM-5.2 Unsafe, GPT-5.5 Wins as the All-Rounder

GPT-5.5 passed 28 of 28, GLM-5.2 passed 26 at an eighth of the cost, and Fable 5 refused nine tasks — four of them entirely benign. Our data, and a deep dive into the refusals.

Read more

The UK AI Policy Landscape: What Enterprise Leaders Need to Track in 2026

There is no UK AI Act, and none is coming. What exists instead is an estate — regulators, compute money, sandboxes, institutes — that only makes sense viewed whole. The map, with the moving parts marked.

Read more

AI Growth Labs: What the UK's Regulatory Sandboxes Mean for Your Sector

Regulatory sandboxes for AI open to legal services this summer, with other sectors to follow. What direct access to your regulator buys, and what to do before applications open.

Read more

The Agent Governance Gap: $202m Budgets, 26 Percent Cost Visibility

Q2 2026 data shows agent deployment plateauing, orchestration doubling and employee resistance quadrupling — with the money committed before the controls exist.

Read more

Is Claude Fable 5 Worth It for Enterprise Coding?

Fable 5 leads every coding benchmark here by a wide margin, at double the price of Anthropic's own Opus 4.8. Whether the premium is worth it, and where GLM-5.2 does the job just as well.

Read more

How to Sandbox AI Agent Code: Firecracker, OpenSandbox, Docker, SmolVM and nono Compared

The brief said microVMs, but only two of the five are. A decision guide to the four isolation models for running AI agent code — and how to pick the right one.

Read more

Multi-Agent Orchestration Frameworks Compared: ruflo, Aperant, Sandcastle, Mission Control and Maestro

An evidence-based comparison of five multi-agent orchestration tools — ruflo, Aperant, Sandcastle, Mission Control and Maestro — by popularity, use case, hosting and community feedback.

Read more

GSD, BMAD, OpenSpec, or GitHub Spec Kit: Choosing the Right AI Development Framework

Four spec-driven AI development frameworks have emerged as the serious options for structured AI coding. They share a starting principle — and diverge sharply on everything else.

Read more

Claude Memory and Dream: Evidence, Architecture and Risk

Reported deployment gains, system architecture, and the governance risks of persistent, self-editing agents.

Read more

GPT-5.5 and the Age of Autonomous Task Completion

OpenAI's GPT-5.5 is the first mainstream model designed for AI that acts rather than assists. Here is what the governance and workflow implications mean for enterprise teams.

Read more

UK AI Regulation Tightens: What the ICO Guidance and Parliament's Inquiry Mean for Employers

The ICO, Parliament, and the DRCF are reshaping what UK employers must do when AI touches decisions that affect people. Here is what has changed and what to do about it.

Read more

The Enterprise AI Reality Check: Why 79% of Organisations Aren't Seeing ROI

AI funding hit $300 billion in Q1 2026. Meanwhile, new research shows 79% of organisations face significant adoption challenges. Here is what the successful 29% are doing differently.

Read more

UK Sovereign AI: What the Government's £500m Bet Means for Your Organisation

Liz Kendall's £500m Sovereign AI programme frames compute concentration as a national security risk. Here is what it signals for enterprise technology strategy and procurement.

Read more

A2A Protocol: An Enterprise Architecture Decision Guide

What A2A standardises, where interoperability still stops, and how to evaluate vendor support in architecture and procurement.

Read more

Global AI Adoption: How Different Countries Are Embracing Artificial Intelligence

From US velocity to EU caution, Chinese state strategy to Middle Eastern sovereign ambition — a country-by-country breakdown of AI attitudes, intensity and what it means for global business.

Read more

Claude Mythos and Project Glasswing: What the Security Evidence Shows

Reported vulnerability findings, limits of the available data, and implications for security teams.

Read more

UK AI Companies: A Field Guide to Ten Organisations

An unranked selection used to examine what the UK ecosystem produces and where it remains constrained.

Read more

RAG vs GraphRAG: Which Retrieval Architecture Is Right for Your AI Application?

Retrieval-augmented generation transformed enterprise AI. Now GraphRAG is challenging the assumptions it was built on.

Read more

Copilot Studio vs Azure AI Foundry vs AWS Bedrock

Three platforms, three layers of the AI stack. A practical guide to choosing the right enterprise AI platform for your organisation.

Read more

The UK's AI Skills Gap: Why Closing It Is One of the Biggest Opportunities of the Decade

73% of UK adults have had no AI training at all. Here's where the opportunity lies — and where to start.

Read more

GitHub Copilot vs Claude Code vs Cursor

A practical comparison of three AI coding tools — with real-world pricing models, costs at scale, and a decision guide for engineering teams in 2026.

Read more

The UK's Public AI Infrastructure: Government Institutes Shaping the National AI Ecosystem

Behind the UK's AI startups and research labs sits a layer of publicly funded institutions most people rarely encounter — from the Alan Turing Institute to the AI Security Institute.

Read more

Microsoft Foundry: Architecture, Governance and Trade-offs

A decision guide to Foundry's control plane, model and agent services, Azure advantages, operational costs and trade-offs.

Read more

Generative AI Adoption by Industry: Use Cases, Data and What Comes Next

A sector-by-sector breakdown of where generative AI is actually being used — from financial services and healthcare to charities and education.

Read more

AI-Assisted Web Development: Where It Helps and Where It Fails

Where coding and content tools save time, where they create risk, and which review controls matter.

Read more