Netlify Agent Runners let you describe an app or a change in plain language and have an AI coding agent build it inside your Netlify project, with a preview you can review before anything goes live. You pick an agent (Claude Code, OpenAI Codex, Google Antigravity, Meta Muse or OpenCode), type what you want, and Netlify runs the agent with access to your project's settings and deploy history. You don't need a technical background to start, per Netlify's docs. It costs credits, and the free plan's 300 credits can go fast. Prices and limits below are from Netlify's own pages, checked on November 20, 2026.
If you saw "Agent Runners" in a Netlify announcement or in your dashboard and wondered what it does, what it costs and what it can't do yet, this guide covers it in plain terms.
What are Netlify Agent Runners?
Netlify is a service that hosts websites and web apps. Agent Runners is its built in way to ask an AI agent to work on one of them. Netlify's docs say you can use it to create a new project from your own prompt or a starter prompt, fix and update code, ship changes and ask questions about your project, all from the Netlify dashboard (Agent Runners docs).
The pitch is context. Because the agent runs on Netlify, it can see your project's environment variables, build settings and deploy history, and it can help check your domains, failed deploys, form submissions and logs without any extra setup. A few details from the docs:
- Two modes. Build (the default) lets the agent change your code and creates a Deploy Preview you can review. Ask answers questions about your project and changes nothing.
- The agent can ask you questions. If your request leaves a decision open, it pauses and asks before it builds.
- A separate branch by default. An agent works on a new branch created off your production branch, so you choose when to publish.
- Rollback and staging. The docs list staging environments, inline Deploy Previews and rollback among the features.
What can you build with it?
Netlify says Agent Runners are best at well defined work: quick fixes, content updates from a marketing or design team, new landing or 404 pages, background chores like fixing broken links and redirects, and changes you make from your phone. It also covers whole new apps. In its announcement, Netlify gives the example of a customer lead capture site where you work on the form, view submissions and check uploaded files from one place (Inside the new Netlify).
Netlify also says its users have changed. In that post the company writes that most people arriving in the last year weren't professional developers, but experts in finance, marketing, retail and other fields. It reports that more than 500,000 users made more than 1.7 million agent runs, and that agent work produced more than 700,000 production deploys in a 30 day window. Those are Netlify's own figures, and I couldn't check them independently.
What are Connectors?
Connectors let the agent pull information from services you already use while it builds. Netlify's docs say it supports over 150 of them, including Notion, Linear, Stripe, Snowflake, Salesforce, HubSpot, Slack and GitHub (Build with connectors). Some example prompts from the docs:
- "Read Linear issue WEB-142 and build the landing page it describes."
- "Review the spec in our Notion page and implement the feature."
- "Create a product at $150 and build a checkout page for it" (with Stripe connected).
Three limits are worth knowing. First, connections are personal. Netlify's announcement says that joining the same project doesn't give you your teammates' connected accounts. Second, connectors work while the agent is building. Netlify says they don't give the deployed app ongoing access, and that this will come soon. Third, an agent can only do what the account you connect allows, so use an account with the right permissions and no more. Connectors launched on November 17, 2026, per Netlify's changelog.
How much do Netlify Agent Runners cost?
Agent Runners run on Netlify's credit based plans. Here are the plans from the pricing page, checked November 20, 2026 (Netlify pricing):
| Plan | Price | Credits |
|---|---|---|
| Free | $0 | 300 credit limit |
| Personal | $9 per month | 1,000 |
| Pro | $20 per month, unlimited members | From 3,000 (tiers up to 20,000) |
| Enterprise | Custom | Custom |
Extra credits come in packs: 500 credits for $5 on Personal and 1,500 for $10 on Pro. Netlify's docs say an agent run uses credits for AI inference (the model's work) and for compute, and that every $1 of model usage converts to 180 credits (pricing for AI features). Publishing a production deploy costs 15 credits on top. On Enterprise, the docs tell you to contact your account manager for Agent Runners.
My arithmetic, not Netlify's: 300 credits divided by 180 is about $1.67 of model usage, before compute and deploys take their share. Netlify's docs warn that on a free plan a single prompt can use most of your credits when you pick an expensive model. The docs rank models from 1 to 5 by cost and suggest trying the same prompt on a cheaper model first. If your team runs out mid run, Netlify keeps the progress and cancels the run, and you can continue once credits return. You can also set an AI credit usage limit to cap spending.
What are the limits?
Before you build your plan around this, check the fine print:
- GitHub only for Git. If your project deploys from a Git provider, Agent Runners work only with GitHub, not GitLab, Bitbucket, Azure DevOps or Cursor Origin.
- No Split Testing. You must turn it off before using Agent Runners.
- Active runs are capped by your plan.
- Requirements. You need a credit based plan, a role such as Team Owner, Developer, Internal Builder or Publisher, and AI features turned on, which they are by default.
- The new experience is rolling out gradually. Netlify says the redesigned dashboard began reaching most new users on November 18, 2026, and existing customers join a waitlist (changelog).
- Netlify Server is early access. The long running backend option is free during early access, with up to 3 instances (1 on Free and Personal) that scale to zero after 5 minutes without requests. Netlify says it will bill for it after early access, at rates it hasn't announced.
On models and privacy, the docs say OpenCode runs through OpenRouter, with routing only to providers that have a zero data retention policy. The other four agents run on models from Anthropic, OpenAI, Google and Meta, and are not routed through OpenRouter.
Should you trust what an agent builds?
Treat an agent like a fast junior teammate: useful, and in need of review. Preview before you publish, ask for plain language explanations of what changed, and never assume an app is secure because it works. Our guide to vibe coding shows where that goes wrong, including a security firm's check of apps built on the Lovable app builder. The same habits apply on Netlify, even with its previews and rollback.
Go deeper
- What is vibe coding: where building by prompt works, where it breaks and the review habit that keeps it safe.
- AI tools for developers: the 2026 tools, sorted by what they do.
The skill behind it
Tools like this one get new features every week. What stays useful is knowing how to describe the result you want, how to check what an agent did and when to ask a person for help. If you'd like to build that skill for your own work, you can request information below.
Sources, checked November 20, 2026: Agent Runners docs, Build with connectors, Pricing for AI features, Netlify pricing, Netlify changelog, Inside the new Netlify and Netlify Server docs.
