Dane Knecht says Cloudflare's AI-first restructuring was hard, but it forced teams to rethink workflows rather than just speed up old ones. A product-manager email task that involved three teams became one tool, and employees took ownership bottom-up based on initiative and passion. He credits that distributed energy for the progress.
Dane Knecht thinks asking humans to choose a model for every task is silly; intelligent routing should match frontier models to planning and cheaper models to execution. Cloudflare grades people on output and sends usage notices rather than maintaining a token leaderboard. He says people are usually happy to get the notices.
Dane Knecht argues cyber capabilities will end up open because attackers will obtain them anyway. Keeping them only in closed frontier models mainly excludes defenders who need access, without stopping bad actors; top companies will still find a way to get them. He sees that as a broad access problem.
Dane Knecht says Cloudflare used to see Log4j-style incidents once a year, but now serious vulnerabilities surface weekly, some as severe though unreported. He expects over half of engineering time to shift to security for roughly 18 months, then productivity should recover as AI scans code.
Asked whether critical infrastructure should be air-gapped, Dane Knecht argues humans adapt quickly and competitors collaborate during security crises. He points to the industry's move from castle-and-moat defenses to zero trust over the past five to ten years as making companies better prepared rather than doomed.
Dane Knecht says the first security principle is shrinking attack surface by simplifying software and writing less code. But early AI models tend to be verbose and add complexity, so humans must enforce simpler designs and adversarial reviews while agents still help at each stage.
Dane Knecht plans to grow Cloudflare from one red team to four or five, focusing on vulnerability discovery. He trusts crisis-driven scale to force engineers to automate patching, because normal processes improve when volume overwhelms them. He says engineers are 'lazy in a good way' and will automate away the pain.
Dane Knecht describes Cloudflare OS as an internal platform that lets employees use AI without leaking data, while teaching agents company-specific practices. It is also a reference architecture for building a company OS on Cloudflare's developer platform, so security is built in rather than added later.
Dane Knecht says recent major vulnerabilities were found by high schoolers pointing agents at targets, not only nation states. He calls them 'agent kiddies,' and bug bounties reward them, so capabilities once reserved for well-funded teams are now available to the masses. That spreads powerful tooling broadly.
Dane Knecht calls this the messy middle: agents crawl a human web built for ads and page views, which breaks the economic model. He expects machine-readable content, new payment rails, and micro-payments so agents can pay a site's CPM rate directly instead of forcing translation layers.
Asked how small organizations without red teams can protect themselves, Dane Knecht says put infrastructure behind Cloudflare. He notes Cloudflare gives major vulnerability protections to free customers too, viewing security as a utility, and points to CDN access broadening from roughly 3,000 sites to hundreds of millions.
