https://xitter.cf/t.co/J7SXybFJEj
Mexico
Joined April 2009
- Tweets28.1K
- Following758
- Followers573
- Likes5.7K
Ramp just opened early access to Inspect, its in-house cloud coding agent. The stat that jumped out: Inspect now accounts for 74% of Ramp's merged PRs, and 99% of its engineering, design and product teams used it in the past month.
ramp.com/blog/introducing-in…
How it works: tag @inspect from Slack, GitHub, Linear or the web, and it turns the request into a visually validated, reviewed pull request using the team's own dev environment. It runs in the cloud, so nobody sets up a local environment first.
Devs ready to let AI write app logic: 18%→50% in a year. Tests 50→75%, refactoring 38→60%, code review 33→43%. Debugging holdout: 54% still won't.
JetBrains Research: blog.jetbrains.com/research/…
From Mexico: Metis is an open-source AI security code review tool from Arm's product security team. No hardcoded rules, just LLM reasoning. It also validates SAST findings to cut false positives, and runs on OpenAI, Anthropic, Ollama and more. 888 stars.
github.com/arm/metis
GitHub made assisted permissions GA in Copilot CLI, the Copilot app and VS Code. When the agent wants to run a command, touch a path or open a URL, a model checks it. Clearly safe stuff runs; the rest still waits for you. Fewer clicks. Still watch it. github.blog/changelog/2026-1…
Claude Code 65.5% and GitHub Copilot 58.7% lead AI coding agents in the Stack Overflow Developer Survey 2026 (n=12,255). Codex 29.5%, Cursor 25.5%. 73% use them daily; 48% trust output only when easily verified.
Source: survey.stackoverflow.co/2026…
Claude Code 65.5% and GitHub Copilot 58.7% lead AI coding agents in the Stack Overflow Developer Survey 2026 (n=12,255). Codex 29.5%, Cursor 25.5%. 73% use them daily; 48% trust output only when easily verified.
Source: survey.stackoverflow.co/2026…
From Mexico: Skylos v5.1.1 shipped today. Apache-2.0 CLI and PR gate for Python: dead code, security bugs, secrets, vulnerable deps and AI-code mistakes. skylos done checks if a coding agent's branch is ready to merge, flagging skipped tests. 852 stars.
github.com/duriantaco/skylos
Prisma shipped an official ChatGPT plugin today. It hooks ChatGPT and Codex up to Prisma ORM, Postgres and Compute, so one prompt can write the data layer, set up the database and deploy. No more hopping between dashboards and CLIs. Review what ships. prisma.io/blog/one-chatgpt-p…
Google Cloud announced Agent Compute Engine on Oct 22. Per Google, it gives each agent its own hardware-isolated microVM that boots fast, pauses while the agent thinks and resumes from snapshots. It already powers Gemini. Preview in the coming months. cloud.google.com/blog/topics…
From Mexico: kubernetes-mcp-server v0.0.69 is out. Apache-2.0 MCP server that lets AI agents work with Kubernetes and OpenShift, Helm included. Go-native: it talks to the API server directly, not a kubectl wrapper. Single binary. 2,180 stars.
github.com/containers/kubern…
GitHub shipped a new GitHub app for Microsoft Teams on Oct 23. Per GitHub, a bug report lands in a chat, you @mention GitHub, and Copilot investigates, fixes it and opens a PR. Your team still reviews and merges. The legacy app retires Nov 16. github.blog/changelog/2026-1…
From Mexico: Crit 0.23.1 is out. It's an MIT, local-first tool for reviewing your coding agent's plans and diffs in the browser. New: a Markdown Source view, and comments survive concurrent writes. 1,241 stars.
github.com/tomasz-tomczyk/cr…
Replit launched Monitoring Agent on Oct 22. Per Replit, it investigates a production alert, validates a fix and emails you a summary. Auto-fix and Auto-republish are separate opt-ins, and republish is experimental. My default: let it fix, I press publish. replit.com/blog/replit-monit…
From Mexico: Google open-sourced the Elephant-Goldfish Model plugin (Apache 2.0), the design-first AI workflow thousands of Googlers use daily to ship production code. The rule: no code until a fresh AI session can explain your design back to you.
opensource.googleblog.com/20…
The Elephant is one long, context-rich session that only does design: it interviews you, debates trade-offs and drafts a Markdown spec down to the exact files to change. Goldfish are zero-context sessions that review it, because they share none of the Elephant's blind spots.
From Mexico: LayerFS shipped v1.0.0 on Oct 21, its first official release. Each AI agent mounts a full writable workspace over shared committed history, commits what changed as deduplicated checkpoints, then branches, rewinds or merges. MIT, 1885 stars.
github.com/Ephemeral-AI-Lab/…
Glean made headless agent building GA on Oct 21. You define an agent as spec.yaml plus instructions.md from Claude Code, Cursor or Codex, test it, review it in Git, then publish it to Glean's runtime, per Glean. Agents are turning into regular code. glean.com/blog/glean-headles…
