Short Stack #11: Replit Cloud Agents let anyone on your team send tested pull requests to the real codebase

Get 5 AI picks like these every weekday at 7am MT. Free, plus 50 vibe-coding prompts when you confirm.

Short Stack #11Wed, Oct 28, 2026

Image from replit.com: Replit Cloud Agents launch card

Replit Cloud Agents let anyone on your team send tested pull requests to the real codebase

Replit launched Cloud Agents for existing repos on Oct 27, for all Pro and Enterprise customers. Describe a change in plain language and the agent edits your company's code, clicks through the running app in a sandbox, then opens a GitHub PR or GitLab MR with screenshots and video attached. You can also start a change by mentioning @Replit in Slack.

⏱ 10-MIN TRY

Connect one repo at replit.com/repos and ask for a small UI fix you've been putting off. Is the PR video enough to review it?

Source: replit.com

Image from canva.dev: Canva MCP developer portal screenshot

Canva MCP is now open to any developer, no waitlist

As of Oct 27, you can create an app in Canva's Developer Portal, turn on Canva MCP and connect to its remote server without applying first. An AI assistant can then generate designs from a brief, edit existing ones and export to PDF, PNG, PPTX or MP4. Each user signs in with their own Canva account, and there's no service-account option.

⏱ 10-MIN TRY

Point your MCP client at https://api.canva.com/connect/v1/mcp, list its tools, then ask for one social post from a short brief and export it as PNG.

Source: canva.dev

GitHub Copilot agents can now run as your organization, not as you

GitHub added server-to-server authentication for the Copilot SDK in public preview on Oct 27. Give a GitHub App the read-only Copilot Requests permission, install it on an org and pass a short-lived installation token. Usage is billed to the org and follows its model policies and budgets, so a bot no longer depends on one person's login.

⏱ 10-MIN TRY

List every automation running on your personal Copilot or API key. Pick the one that would break if you left, and note what an org-owned app would need.

Source: github.blog

Illustration: a browser window with a small chip inside it and a crossed-out cloud

Opera's WebAI lets websites run AI models on the visitor's own device

Opera introduced WebAI on Oct 27, a proposed web standard for running open-weight models like Gemma inside the browser, plus Apple's on-device model on Macs. It's an opt-in early preview in Opera One. Opera's pitch: no per-request AI bills, and data stays local. Speed depends on the user's hardware.

⏱ 10-MIN TRY

Find one AI feature in your product that only classifies or summarizes short text. Estimate what it costs per month, the number a local model would save.

Source: techcrunch.com

Illustration: a tangled puzzle grid with a shortcut line cutting through it

OpenAI and Anthropic both claim their models refuted a core computer science assumption

OpenAI released AI-generated work it says refutes the Exponential Time Hypothesis, a decades-old assumption about how hard certain logic puzzles are to solve. Hours later, Anthropic released its own refutation. Neither is peer-reviewed. MIT's Ryan Williams, who is organizing the field's review, told TechCrunch OpenAI said it checked and was reasonably confident it doesn't break any crypto systems.

⏱ 10-MIN TRY

Ask your favorite model to explain the Exponential Time Hypothesis in five sentences, then ask what would change for a developer if it's false.

Source: techcrunch.com

Replit Cloud Agents let anyone on your team send tested pull requests to the real codebase

Canva MCP is now open to any developer, no waitlist

GitHub Copilot agents can now run as your organization, not as you

Opera's WebAI lets websites run AI models on the visitor's own device

OpenAI and Anthropic both claim their models refuted a core computer science assumption

Get tomorrow's in your inbox.