In this episode, host João Tomé is joined by Celso Martinho, VP of Engineering at Cloudflare, to discuss two major launches: Markdown for Agents and Moltworker (for OpenClaw) — and what they signal about the future of AI agents on the Internet.

Celso explains how Markdown for Agents was conceived, built, and shipped in just one week, why AI systems prefer markdown over HTML, and how converting a typical blog post from 16,000 HTML tokens to roughly 3,000 markdown tokens can reduce cost, improve speed, and increase accuracy for AI models. We also explore Moltworker, a proof-of-concept showing how a personal AI agent originally designed to run on a Mac Mini can instead run on Cloudflare’s global network using Workers, R2, Browser Rendering, AI Gateway, and Zero Trust.

We discuss observability for AI crawlers, new monetization models for publishers, the rapid growth of agent ecosystems, and why AI is becoming less hype and more infrastructure.

Mentioned blog posts:

• Introducing Markdown for Agents
• Introducing Moltworker: a self-hosted personal AI agent, minus the minis

⏱️ Timestamps

1:15 — Introducing Markdown for Agents

1:46 — From idea to ship in one week

2:37 — Why AI systems prefer markdown over HTML

3:30 — HTML “packaging” vs semantic content

4:39 — How Cloudflare converts HTML to markdown in real time

5:19 — Token savings: 16,000 vs 3,000 tokens

6:29 — Context windows, cost, and AI efficiency

8:21 — Tracking markdown trends in Cloudflare Radar

9:05 — Live demo: content negotiation header with curl

11:07 — AI projects in Lisbon: AI Search, PaperCrawl, and more

12:36 — Observability and new monetization models for publishers

13:56 — What is OpenClaw and why it went viral

14:54 — From Hacker News to Cloudflare in hours

17:06 — Running OpenClaw on Cloudflare instead of a Mac Mini

18:05 — Why this is a proof of concept (not a product)

20:06 — Architecture: Zero Trust, Workers, R2, Browser Rendering, AI Gateway

22:32 — Demo: AI agent records and posts a video automatically

24:53 — 10,000 GitHub stars and open source support

26:11 — AI in 2026: intensifying work, not replacing it

In this episode of This Week in NET, host João Tomé is joined by Emily Hancock, Cloudflare’s Chief Privacy Officer and Data Protection Officer, for a wide-ranging conversation about privacy in 2026 and how the role has evolved in the age of AI.

Emily explains how privacy officers shifted from GDPR compliance to broader data governance, responsible AI practices, cybersecurity collaboration, and cross-border data frameworks. We explore privacy by design, data minimization, vendor risk, government requests, warrant canaries, digital sovereignty, insider threats, and how AI is reshaping both attacker and defender capabilities.

We also discuss Cloudflare’s approach to responsible AI, how teams use internal controls to avoid misuse of customer data, and why “human in the loop” remains essential for accuracy, safety, and trust.

Check the Cloudflare Blog: blog.cloudflare.com

1:53 — Blogs roundup

3:58 — How the CPO role has evolved since GDPR

7:04 — From GDPR to AI governance

9:46 — Privacy + cybersecurity: breaches, notifications, preparedness

14:08 — “Fire doors” and incident containment

14:56 — Privacy by design & data minimization

20:07 — Government requests, due process, and transparency

22:08 — Warrant canaries & what Cloudflare will never do

23:17 — Digital sovereignty: localization and global differences

26:25 — Data Localization Suite & Metadata Boundary

28:06 — AI and privacy: rules, training, customer protections

29:35 — Cloudflare’s AI principles

31:32 — AI sovereignty & running inference close to users

32:19 — “AI as an intern”: accuracy and human review

34:31 — Protecting personal data when using AI

36:20 — What’s coming in 2026: regulation & fragmentation

38:37 — Insider threats & Zero Trust

40:33 — Emily’s privacy wish list for 2026

In this episode, David Belson — Cloudflare’s Head of Data Insights — joins us to walk through the biggest Internet disruptions of late 2025 and early 2026.

At the start, we also highlight several new posts on the Cloudflare Blog: Moltworker, a self-hosted personal AI agent built with OpenClaw (former MoltBot and ClawdBot) and Cloudflare’s Developer Platform; Post-Quantum Matrix Homeserver, a proof-of-concept encrypted messaging server running entirely on Cloudflare Workers; Route Leak Incident (Jan 22), what happened in Miami and how routing policy safeguards are being improved; Google’s AI Advantage, why crawler separation is needed for fair competition and better protection for publishers.

We then go into the major Internet trends, including the storm-related disruption in three regions in Portugal this week. Our main focus is the government-directed nationwide shutdown in Iran.

Then we also go over Q4 2025 disruptions: repeated weather-driven outages across Africa and the Caribbean, submarine cable failures, DNS anomalies, and the persistent risk of centralized points of failure. David also explains how Starlink’s global footprint is reshaping Radar visibility — and why the Internet remains remarkably resilient despite a turbulent quarter.

Mentioned blog posts:

• Cable cuts, storms, and DNS: a look at Internet disruptions in Q4 2025
• Introducing Moltworker: a self-hosted personal AI agent, minus the minis
• Route leak incident on January 22, 2026
• Building a serverless, post-quantum Matrix homeserver

⏱️ Timestamps

0:30 — Weekly blog roundup (Moltworker, Route Leak, Google’s AI Advantage)

4:13 — Storm impact in Portugal: what Radar saw in Leiria, Santarém, and Coimbra

11:55 — Iran’s multi-week Internet shutdown: scale, signals, and how it unfolded

18:15 — The “National Information Network”: partial access, allowlisting, and blocked services

21:24 — Power vs. connectivity: how electricity failures show up as Internet outages

22:33 — Q4 global round-up: Jamaica, Sri Lanka, Indonesia, and cyclone-driven disruptions

30:32 — Technical failures: ISP issues, DNS problems, routing mistakes, and what Radar detects

33:47 — The future of Radar: Starlink visibility, provider-level metrics, and disruption heat maps