|
buildwith.sh // issue 006 // Nov 19, 2026
27 new drops since last Thursday. Replit will run a swarm of agents toward an outcome on a budget you cap, Vercel shipped eve 1.0, and GPT-6.2 landed in Copilot. Here are the eight worth your time.
$ ls ./new --since=last-issue --top=8
Replit Missions [Nov 17 · prompt-to-app] You set an outcome, a spending cap and which actions need your sign-off, and Replit runs a swarm of cloud agents that plan, build, test and iterate toward it, sending updates on the schedule you pick. An independent AI judge checks results against the success criteria you agreed on; if the budget runs out first, work stops.
eve 1.0 by Vercel [Nov 17 · orchestrate] Vercel’s open-source framework for building production agents is now generally available. You own the harness: pick the model, add or replace tools, edit instructions and choose which actions need approval, and the Workflow SDK lets an agent resume from its last completed step after a crash.
GPT-6.2 Astra and GPT-6.2 Sol in GitHub Copilot [Nov 17 · agent] OpenAI’s newest coding models are generally available in Copilot for Pro+, Max, Business and Enterprise. Astra is for agentic coding and terminal work; Sol is for complex agentic coding and in-depth code review.
Gemini 4 Hydrogen by Google [Nov 12 · agent] Google's new fast model for everyday coding and agentic work: repo coding and debugging, interactive web development and multimodal tasks, with four thinking levels. It can run alone or alongside Gemini 4 Carbon.
Claude Code 2.1.327 [Nov 17 · agent] Mostly safety and reliability fixes, plus claude mcp add-json for keeping tokens off the command line.
ngrok MCP server [Nov 16 · deploy] ngrok now has a remote MCP server on every plan, free included. Your coding agent can list and explain endpoints and domains, query HTTP and TCP traffic from Traffic Inspector, write and validate Traffic Policy rules (rate limits, blocks, headers) and save them to cloud endpoints, and search the ngrok docs.
Antelog [Nov 14 · orchestrate] A safety layer that checks every command your coding agent wants to run and stops the ones that would wipe production data or backups, tear down cloud resources, force-push over main, or leak secrets and customer data, including prompt-injected sends.
Burnstate [Nov 14 · orchestrate] Mac menu bar app that shows your Claude Code, Codex, Cursor and Vercel AI Gateway limits against the clock and forecasts when you'll run out before the reset. Also answers Claude Code permission prompts from a card and lists every session's tokens and estimated cost from local logs. No account or analytics.
All 27 drops, newest first: buildwith.sh/new · RSS
$ diff rankings --since=last-week
+ Devin pricing re-verified Wednesday's pass confirmed Devin's team plan at $80/month plus $40 per full seat, up to 100 users, and noted that its pricing page and billing docs describe quotas differently.
+ 60 tools re-checked Links, pricing and details were refreshed across the board on Nov 18. Rankings stay evidence-based and are never paid.
Rankings are re-checked every Wednesday and never paid. See the board.
$ cat tip.md
Keep tokens off the command line. Claude Code 2.1.327 adds `claude mcp add-json <name>`, which reads the server JSON from stdin, so API tokens never land in your shell history or process list. Pipe it from your password manager or a secrets file instead of pasting keys inline. Also new: /status now lists broken agent, skill and command files and hooks that never match. More in Agent Ops.
$ cat sponsor.txt
This slot is open. Founding Sponsor weeks start at $99 and include a sitewide bar plus a line in this newsletter. Book a week.
|