What I'm building.
Crox isn't just a consultancy. It's a place where I research, build, and write about how AI actually works inside regulated businesses. Here's what's public.
May 2026
Static Controls, Live Models
A briefing for boards and exec teams adopting AI in regulated industries. Free to read. No email gate. Free to forward.
Read the briefing →Product
Canary
Automated CQC monitoring for the UK care sector. Free email alerts on inspection changes. Pro tier adds AI-powered analysis, multi-area monitoring, and a monthly national brief. A working example of AI applied where the stakes are real.
Live at canary.crox.io →Product
BBX
A black-box recorder for AI systems. Every request and response is captured and stored for auditability — the flight-recorder pattern applied to AI, so governance questions get answered from evidence rather than memory.
Live at bbx.crox.io →Product
Radar
Early visibility for care homes. Frontline observations are captured as weak signals, grouped into themes, and tracked as trends — so patterns surface before they become incidents. Deliberately calm: no scores, no traffic lights.
Live at radar.crox.io →Tool
NEWS2 Calculator
The NHS National Early Warning Score 2, as a quick web calculator — enter observations, get the score and escalation guidance. Built to NHS England and NICE guidance for clinical deterioration.
Live at news2.crox.io →Product
Fred
An AI agent for families — the assistant patterns we build for businesses, pointed at the admin of family life instead.
Live at fred.crox.io →Assistant
Adam’s AI Assistant
The AI assistant on this site — bottom right of every page. Answers questions about Crox, qualifies enquiries, and books calls. The same pattern we build into clients’ sites: a real assistant with guardrails, not a chat toy.
Live on crox.io →Experiment
Common Ground
An anonymous social game: climb ten levels of purgatory by finding common ground with strangers — three matching answers to advance, one mismatch and you both fall. Real-time matchmaking, retro CRT aesthetic, built for the joy of it.
Live at common-ground.crox.io →Experiment
Attack of the Frog Army
A silly browser game. Exists because not everything needs a business case — and because shipping small things fast is a habit worth keeping.
Live at frog-army.crox.io →Experiment
Party Parrot Yourself
Upload a photo, trace your head, download yourself as a partying parrot GIF. Vanilla JavaScript, Web Workers for the GIF rendering, zero business case.
Live at party-parrot.crox.io →Research
DODAR Research Site
The open home of the DODAR structured-reasoning work: benchmark results across 8 LLMs showing small models matching frontier quality at a fraction of the cost, with the method published in full.
Live at dodar.crox.io →Directory
MCP Servers Directory
150+ integrations between AI assistants and business tools — Xero, HubSpot, Salesforce, Slack, GitHub, more. Categorised and searchable. Useful when someone asks "can I connect AI to [our tool]?"
150+ servers · categorised →Benchmark Study: Does Reasoning Structure Shape Failure, Not Just Accuracy?
1,500 benchmark tests show DODAR-based prompting doesn't improve accuracy — but it measurably changes how models fail (p = 0.0003).
We Gave Cheap AI Models a Pilot's Decision Framework. They Outperformed the Best.
How the DODAR structured reasoning pipeline enables small language models to match or exceed frontier model quality at 89% lower cost.
AI Governance for SMEs: What You Actually Need Before You Deploy
Not a framework. Four things on two sides of paper — an inventory, a never-list, a named human, and what happens when it is wrong.
AI for GP Practices: The Document Mountain, DNAs, and the Lines Not to Cross
Document workflow, recalls, and care navigation support — where AI genuinely helps general practice, inside NHS information governance rather than around it.
AI for Logistics and Haulage: PODs, Chasers, and Cash Flow
POD chasing, invoice matching, status queries, and load quoting — the transport office is one of the best-fit environments for AI there is.
AI for Hospitality: Win the Function Enquiry, Answer Every Review
Event enquiries answered at midnight, reviews replied to in your voice, allergen paperwork that keeps itself current — where AI pays back in restaurants, hotels, and venues.
AI for Architects: Fee Proposals, Planning Packs, and the Drawing Line
The written layer of practice — proposals, planning documents, minutes, RFIs — automates well. The architecture stays yours, and the profession's rules make the boundary easy.
AI for Construction Firms and Trades: Quotes, RAMS, and Getting Paid for Variations
Faster quotes, job-specific health and safety paperwork, and site notes that become records — where AI genuinely helps a UK construction firm, and what to keep away from it.
AI for SME Manufacturers: Skip the Robots, Fix the Office First
RFQs, order entry, quality paperwork, and supplier chasing — the office layer around the shop floor is where AI pays back for a 20–200 person manufacturer.
AI for Recruitment Agencies: The Wins, and the One Area the Law Watches
Job ads, call notes, CRM hygiene, and candidate comms automate beautifully. Screening is where the EU AI Act and UK GDPR are looking — here's how to draw the line.
AI for Estate Agents: More Viewings, Fewer Evenings Writing Listings
Listing copy, enquiry triage, vendor updates, and compliance checklists — where AI genuinely pays back in a UK agency branch, with honest costs.
AI for Dental Practices: Filling Chairs Without Risking Records
Recalls, letters, front-desk FAQ, and treatment plan follow-ups — the operational layer where AI pays back fast, and the governance that keeps the CQC and UK GDPR happy.
EU AI Act: Does It Apply to a UK Business?
The Act follows the market, not the company registration. What went live on 2 August 2026, what moved to December 2027, and how to work out which half reaches you.
Why Fibery, Not Obsidian, for My Second Brain
Notes on why typed entities beat flat files once the goal is thinking, not just capture.
Your AI Vendor's Bad Day Is Your Regulatory Exposure
On platform dependency risk in regulated AI products.
AI for UK Legal Practices: Automation That Stays Compliant
How solicitors and legal firms are using AI for document review, research, and client intake — within SRA and GDPR boundaries.
5 Ways Small Businesses Are Using AI to Save 10+ Hours a Week
Real examples from UK businesses that connected AI to their tools — and the hours they got back.
AI for Accounting Firms: What's Actually Possible in 2026
From automated bank reconciliation to client communication — a realistic look at what AI can and can't do for UK accounting practices right now.
Why Your AI Assistant Can't See Your CRM (And How to Fix It)
The technical gap between ChatGPT and your HubSpot, Salesforce, or Pipedrive — and the new protocol that bridges it.
How to Connect AI to Your Business Tools Without Writing Code
A step-by-step guide to linking ChatGPT, Claude, and Copilot to your CRM, accounting software, and project management tools using Model Context Protocol.
Is ChatGPT Safe for Business Data? The Honest Answer for SMEs
The tool is rarely the problem — the tier and the habits are. Where the risk actually lives, and the five rules that make AI assistants safe for company data.
What Should an SME AI Policy Include? Seven Questions, One Page
The AI policy that works for a 10–200 person firm is a page long and answers seven questions. Here they are, with the wording that holds up.
How Much Does AI Automation Cost a Small Business? Real Numbers
The three tiers of AI spend — off-the-shelf tools, connected automation, and ongoing support — with actual price ranges and the arithmetic for deciding what's worth it.
What Could Go Wrong? Governing AI Automation Without Killing It
The governance questions boards raise — data security, accuracy, human-in-the-loop, accountability — and the controls that let you move fast with your eyes open.
Build, Buy, or Partner? Choosing How to Add AI to Your Operations
Off-the-shelf tools, in-house development, or a specialist partner — how to decide, the total cost and lock-in trade-offs, and a clear decision checklist.
What's the ROI of AI Automation — And How Do We Prove It?
Building the business case execs actually ask for — hard savings versus soft gains, a simple payback framework, and the ROI traps that sink credibility.
Where Should We Start With AI Automation? How to Pick the First Process
A practical guide for operational teams told to 'go do AI' — how to spot the right first process, score the candidates, and start where the risk is lowest.
How AI Assistants Connect to Your Tools: MCP Architecture Explained
A visual walkthrough of clients, servers, and the communication layer that makes AI integration work — from request to response.
MCP vs APIs vs Plugins: Which Integration Approach Should Your Business Use?
A clear comparison of the three ways to connect AI to your business tools — with guidance on when each one makes sense.
What Is the Model Context Protocol? A Plain English Guide
The open standard that lets AI assistants like ChatGPT and Claude securely talk to your business software — explained without jargon.
Curious about any of this?
Pick one and read it, or start a conversation — the chat in the corner knows what I've built and can point you at the right one for what you're trying to do.
Start with a conversation →