From Rural Lincolnshire to the Edge: Four Days at Cloudflare Connect

There's something absurd about flying from rural Lincolnshire to Las Vegas. One minute you're surrounded by fields and measured rhythms, the next you're walking through the Aria's gleaming corridors whilst thousands of engineers converge to discuss infrastructure that powers a significant chunk of the internet. This was Cloudflare's first global Connect event, and if you're going to make a statement about building systems that span continents, Las Vegas provides the appropriate theatre.

Day one happened to fall on Ada Lovelace Day. Whilst engineers filled session rooms to discuss edge computing, I kept thinking about the woman who imagined machines composing music in 1843. Ada Lovelace saw creative potential in computing when everyone else saw arithmetic. That vision felt particularly relevant as we spent the day building AI agents that query real APIs rather than hallucinate statistics. The women speaking and architecting at Connect were doing exactly what Ada did: refusing to accept limits on what's technically possible.

When AI stops guessing

The university day sessions with Kristian Freeman moved fast. Morning covered Workers fundamentals, afternoon tackled external APIs, Workers KV, and D1 databases. What struck me immediately was how Cloudflare has made correct architectural decisions feel inevitable. Writing code that deploys globally in seconds shouldn't feel remarkable, yet it does when you're coming from traditional infrastructure.

Then Craig Dennis's AI workshops changed the frame entirely. The Model Context Protocol isn't just another abstraction layer. We built tool servers that let AI agents interact with external systems reliably. The data.police.uk integration CDS has created demonstrated why this matters. An AI analysing crime patterns can now query the actual API rather than inventing plausible-sounding numbers. When you're building systems people depend on, that distinction is everything.

This pattern kept recurring. AI has moved from summarising documents to orchestrating multi-step workflows with real systems. By Tuesday, Cloudflare announced partnerships with Visa, Mastercard, and American Express for Web Bot Auth, essentially creating digital passports for AI shopping agents. The problem they're solving is urgent: merchants can't currently distinguish between legitimate agents acting on behalf of customers and sophisticated bots running fraud schemes. When these payment networks align on a standard alongside Shopify and Webflow, you're looking at immediate reach across millions of merchants. That's how experimental technology becomes production infrastructure.

Security by prevention

Matthew Prince opened Tuesday's general session with a vision for the Connectivity Cloud that managed to be both ambitious and grounded in production reality. Stephanie Cohen then led customers from Condé Nast, Shopify, OpenAI, and Visa through what they're actually building. These weren't aspirational case studies. These were people solving specific problems at scale.

One phrase stuck with me from an observability session: automation without governance is risk at scale. We're drowning in data without context, automating everything whilst forgetting to build frameworks that make automation safe.

Firewall for AI represents a different approach. It sits between your application and the model, scanning requests before they reach the LLM. Response scanning comes soon. This shifts security from detection to prevention. Requests attempting to leak sensitive information never reach your model. For public sector work, where data protection is legally mandated, this architectural pattern opens possibilities that weren't previously viable.

The photography site migration provided numbers worth noting. They moved from 13 petabytes to 2 petabytes of storage using Super Slurper in under two weeks. Cloudflare Images transforms everything on demand, eliminating multiple stored versions. That's an 85 per cent reduction in storage costs with better performance. These aren't marginal improvements. They change what's economically feasible.

Post-quantum cryptography received serious attention. The message was clear: this isn't future-proofing anymore. Teams are implementing it now. The threat model has shifted from theoretical to imminent.

Production patterns that actually work

Contentful's session on personalisation revealed what Durable Objects handle at scale. Self-healing systems with elegant architectures: redirects for rehoming objects, alarms for compliance-driven data deletion, WebSocket connections for live content syncing. These were production patterns, not conference demos.

Web Fragments provides a systematic way to rebuild massive applications using different technologies. Cloudflare is using it to reconstruct their entire dashboard, which tells you everything about production readiness. For legacy systems needing incremental modernisation rather than risky replacements, this pattern matters.

TanStack Start caught my attention for inverting conventional wisdom. Route loaders are isomorphic by default, running on the server during server-side rendering and client during navigation. This eliminates redundant requests that plague other frameworks. It's philosophically client-first, treating single-page applications as the default with opt-in server-side rendering, rather than the server-first approach of Next.js. For data-intensive applications where client state is primary but you want server-side rendering performance, this makes considerable sense.

The exam processing talk was refreshingly honest. Design for human-in-the-loop because you know it's going to fail. Their infrastructure team's frustration with Wrangler deployments got knowing laughs from an audience recognising the gap between conference demos and production reality.

One insight challenges current practice. Structured outputs for LLMs performed worse across most models they tested. Instead, they use a strong model for unstructured outputs, then a cheaper model to structure it afterwards. That's backwards from what most teams are doing. It's the kind of counterintuitive finding that only emerges from production experience at scale.

What 405 terabits per second actually means

Cloudflare's network operates at 405 terabits per second globally. That's Windows 11 downloaded in under one millisecond. But raw capacity matters less than what it enables. Workers AI demonstrations showed text-to-speech and speech-to-text running at the edge rather than centralised inference. When you push intelligence to where users actually are, latency drops to levels that change what's possible in interfaces.

Edge compute isn't novel. What's changed is the sophistication of what can run there. The combination of powerful developer tools, security by default, and genuine global distribution opens architectural possibilities that weren't feasible two years ago.

What this means for public services

For CDS and public sector projects, several approaches warrant serious investigation. Web Fragments for rebuilding complex legacy applications incrementally. Workflows for integration systems needing human oversight and proper audit trails. Firewall for AI to guarantee sensitive data never leaks through LLM-based services. Post-quantum cryptography implemented before it becomes urgent rather than forward-looking.

These are production architectures being used today by organisations with demanding requirements. The public sector faces particular challenges around resilience, security, and accountability. When you're building services citizens depend on, shifting from detection to prevention matters enormously. When you're managing legacy systems that can't be replaced overnight, incremental modernisation becomes essential.

The developer experience Cloudflare has built sets a new standard. Push code, it deploys globally in seconds, debug it with proper observability tools. That's the bar now, changing expectations for what's reasonable to achieve.

The experience beyond code

The welcome reception at The Hub provided the first chance to feel the event's energy. Sponsor booths, a Knowledge Bar staffed by Cloudflare experts, the Radar Soda Lounge, and a Community Zone that genuinely facilitated conversations where ideas develop.

Tom Evans's partner summit closing highlighted the ecosystem developing around Cloudflare. When evaluating platforms, partner network strength matters as much as underlying technology. Seeing award winners and understanding implementation range across sectors provided useful context.

Then there was the after-dark event. Common and Gwen Stefani performing at a bowling alley shouldn't work, yet it absolutely did. That combination of technical rigour during the day and unabashed entertainment at night captured something essential about running conferences. People remember moments, not just slide decks.

What we're taking back

Flying back to rural Lincolnshire with a notebook full of ideas, the contrast between environments felt sharper. Las Vegas represents a particular technological optimism: bold, loud, unapologetically ambitious. But the problems we solve at CDS require different ambition. Building resilient public services that work for everyone demands technical excellence combined with genuine understanding of how people interact with government.

What Cloudflare demonstrated was that you can build infrastructure operating at global scale whilst maintaining developer experience that feels human. The technical sessions were excellent, but real value came from conversations with product managers and developers who understand the problems we're solving because they've solved similar ones at production scale.

We're already planning which solutions to prototype first. Edge compute, security by default, and genuinely good developer tools open possibilities for how we build public services. Not because the technology is novel, but because it's mature enough to rely on for critical systems.

Four days in Las Vegas provided enough technical insight and practical patterns to influence our work for months. More importantly, it demonstrated what happens when you gather people who care deeply about building things that work reliably at scale. The technical content was superb. Booking Common and Gwen Stefani showed they understand creating moments people remember. Sometimes the audacity of Las Vegas turns out to be exactly the right backdrop.