BuildersCloudflare: post-quantum Privacy Pass and the cost of 2,420 bytesPrivacy Pass protects sites from abuse, but it stands on classical cryptography: a quantum attacker will be able to forge tokens at will. Post-quantum signatures are 2,420 bytes against 64, and MOLE hides who issued the token too.BuildersVercel launches Security Review in beta: it scans only the change on every push, and reviews are free until 10 November 2026Security Review is a new capability in Vercel Agent that looks only at the change on every push to GitHub. It is available to Pro and Enterprise, and reviews are free until 10 November 2026.BuildersCloudflare makes protection for private servers generally available: a DNS record to an internal address, with WAF, caching and Workers running in frontOn 30 October 2026 Cloudflare announced that its services for private origins are generally available (GA). You point a proxied record at a private IP and get WAF, caching and Workers without opening a firewall port to the Internet. The entitlement, though, is switched on by your account team.BuildersGitHub's Copilot Fusion judges what a task needs, and Enterprise gets Copilot without a separate subscriptionAt Universe 2026 GitHub showed Copilot Fusion: by GitHub's own numbers, Fusion Max lands within two points of Opus 5.5 Max at half the cost. For Enterprise, Copilot is no longer a separate purchase. No availability was given for Fusion.BuildersOctoverse 2026: agents now start over 40% of pull requests on GitHub, up from 17% in MarchBy GitHub's data, in September agents started over 40% of pull requests on the platform, against 17% in March. How many of them reach a merge is not stated.BuildersVercel CI is in beta: it runs only what the change touches, and the setup lives in turbo.json instead of YAMLOn 28 October Vercel put Vercel CI in beta on all plans. By Vercel's data, getting from push to green CI is 2-6x faster for them. The announcement gives no price.BuildersReplit's ViBench 2.0: GPT-6.1 Sol leads at 65%, and no open-weight model passes a full buildReplit's new ViBench 2.0 tests an agent from an empty project to a whole product. By Replit's data, GPT-6.1 Sol leads at 65% pass@1, GPT-6 Astra is at 60%, Opus 5.5 at 49%.BuildersMicrosoft Copilot harness in Foundry: the agent loop and the sandbox are Microsoft's, and billing starts on December 1On October 28 Microsoft released the Copilot harness in Foundry Agent Service in public preview: a managed service for agents that reason, use tools and finish long tasks. Harness time is not billed until December 1, then it is billed by consumption, while models and tools are paid separately already.
76 stories