Reinvently
reinvently.co.uk
AI Governance & Compliance
Think Tanks & Policy Research
AI Evaluation & Observability
Claim pageClaim
Follow
Overview
News
Salaries
Products
People
Growth
Offices
Financials
Overview
Reinvently Home News Guides Ed-o-meter Tools About Subscribe Reinvently is an independent AI research organisation. We evaluate models and engineering tools, publish original benchmark data, and analyse the governance and adoption questions shaping how AI is used in the UK.
News
How Much Does AI Improve Software Development Productivity?
A systematic review of 116 studies finds 10-30% higher release output. Agent-native cases report a 4.5 median and an 18 task-specific upper end.
Read more
Report
LLM Benchmark Design: Building the Ed-o-meter
How I built Featherbench, a single-file LLM eval harness running 28 tasks: one variable held constant, refusals counted as failures, and judge bias measured.
Read more
Report
GLM-5.2 vs Fable 5 vs GPT-5.5: 28-Task Eval Results
GPT-5.5 passed 28 of 28 tasks, GLM-5.2 passed 26, and Claude Fable 5 refused to do 5 of them. Full results, costs, and a refusal deep dive.
Read more
Report
The Agent Governance Gap: $202m Budgets, 26 Percent Cost Visibility
KPMG's Q2 2026 pulse: enterprises plan $202m average AI spend but only 26 percent see costs in real time. Why the agent governance gap decides who gets ROI.
Read more
Report
Is Claude Fable 5 Worth It for Enterprise Coding?
Claude Fable 5 tops FrontierSWE at 90 percent but costs $10/$50 per MTok, double Opus 4.8. Whether the premium is worth it, and when GLM-5.2 is enough.
Read more
Report
UK AI Growth Labs: Regulatory Sandboxes
The UK's AI Growth Labs open to legal services first in 2026, with the ICO and 3 legal regulators involved. What regulatory sandboxes mean for your sector.
Read more
Report
UK AI Policy Landscape: What to Track in 2026
No UK AI Act exists. Instead: sector regulators, a £500m sovereign compute programme, AI Growth Labs and a Hardware Plan. The 2026 map for enterprise leaders.
Read more
Report
How to Sandbox AI Agent Code: microVM Options
Firecracker, OpenSandbox, Docker Sandboxes, SmolVM and nono compared: four isolation models for AI agent code, from 5 MiB microVMs to zero-setup sandboxes.
Read more
Report
Multi-Agent Orchestration Frameworks Compared
Compare Multica, ruflo, Aperant, Sandcastle, Mission Control and Maestro by use case, hosting, maturity and GitHub activity.
Read more
Report
How to Choose an AI Development Framework
GSD, BMAD, OpenSpec and GitHub Spec Kit compared: choosing a spec-driven AI development framework, from the 90,000-star Spec Kit to BMAD's $2,000 monthly cost.
Read more
Report
Show more news