While the big headlines and news sites are running away from Network API (except TelecomTV who did a great summit this week) - quietly the boring, but hard work to make the #API’s productive is happening. This week headlines - from UK operators BT Group & Virgin Media O2 finally announcing to launch number verify V2, to Canada and France getting Aduna Global coverage. All of this is get the "boring" fraud to work - but in the mean time building the foundation for the more exciting KYC, QoS/QoD APIs

Machines are ruling the web now

I got a lot of pushback on my post about agents changing the traffic mix. Fair. The big open question: where will the agent run - on the device, or in the cloud? That decides the traffic pattern. I'm still kicking my thinking around.

My instinct - history says local wins where it can. Look at gaming. Cloud gaming makes sense in theory, but the business model won: you already paid for the phone's compute upfront, so it gets used. Same logic for agents. Anything that doesn't need the latest NVIDIA and can run on a small model on-device, will run on-device. What share of agent tasks that covers, I honestly don't know yet.

But whichever way that split lands, agent behaviour is already reshaping traffic - and business models. Bots are now 57.5% of HTML traffic on Cloudflare 's network - humans are the minority (Cloudflare Radar, 2026). As discussed in last newsletter - AI traffic on Fastly grew ~30% between January and May, 6.5x faster than human traffic (Fastly, 9 June 2026). And the stat that gets me: 51% of AI requests go all the way to origin servers, versus under 9% of human requests. Agents don't cache. They fetch.

And they don't give much back. ClaudeBot crawls roughly 24,000 pages for every visit it refers. Perplexity ~111. Old-fashioned Google search: about 5 (Cloudflare Radar). Caveat: app traffic often arrives without referrer headers, so these ratios overstate it - but the direction is not in doubt.

Publishers will need to react. And they're learning what telcos learned twice: traffic over your infrastructure is not the same as value in your pocket.

Two questions I'd love your view on:

  • If agents become the traffic, who pays for the infrastructure? Pay-per-crawl? An agent tier?

  • Does the local vs cloud split actually change the economics — or just who feels the pain first?

The News

Keep reading