How do I make small website edits with AI instead of paying a developer every time?

This page may contain affiliate links from which Blur Studio earns commissions.

Paying a developer's minimum call-out fee to fix a typo or swap a phone number on your site is one of the more quietly annoying costs of running a small business website. It's not that the work is hard, it's that nobody wants to pick up a two-minute job for less than their minimum rate, so you either batch up a list of small fixes and wait, or you pay the fee anyway. Netlify's Agent Runners is built specifically to close that gap: you type what you want changed, an AI agent does the actual work inside your real project, and you get a preview link to check before it ever touches your live site.

How Agent Runners actually works

You open your site in the Netlify dashboard and describe the change you want in plain language, the same way you'd explain it to a person: "change the phone number in the footer to 555-0142," "add a new page listing our three service areas," "make the hero heading bigger on mobile." The agent reads your actual codebase, makes the edit, and deploys it to a private preview URL rather than your live site. You look at the preview, and if it's right, you merge it live; if it's wrong, you either refine the prompt or throw the attempt away with nothing published.

That preview step is the entire safety model here, and it's worth taking seriously rather than skimming past: never approve a change by trusting the agent's description of what it did, actually open the preview link and look at the specific thing you asked for.

Step by step: making your first AI edit

1. Start with something low-stakes. A text change, a new simple page, or a small visual tweak, not your checkout flow or a page handling customer data.
2. Write the instruction like you're briefing a person, not a search engine: name the exact page, the exact text or element, and what it should become.
3. Let the agent run and open the preview link it hands back rather than trusting a text summary of the change.
4. Check the specific thing you asked for on both desktop and mobile widths, since layout prompts in particular can look right on one and wrong on the other.
5. Merge if it's right, or refine the prompt if it's close but not quite, rather than trying to hand-edit the agent's output yourself unless you're comfortable in the code.
6. Set a credit limit before you hand this to anyone else on the team. Team Owners can cap AI credit usage, which matters because an agent that gets stuck on a hard problem can keep retrying and burning credits with nothing to show for it.

What this actually costs

Agent Runners runs on the same account-wide credits as the rest of Netlify, not a separate subscription. Netlify's current plans are Free ($0, 300 credits), Personal ($9/month, 1,000 credits), Pro ($20/month, 3,000 credits) and custom-priced Enterprise, confirmed directly on Netlify's own pricing page. Within that pool, production deploys cost 15 credits each, compute runs 10 credits per GB-hour, and bandwidth is 20 credits per GB; AI inference for Agent Runners is priced per model rather than a flat per-run rate, so a small text edit on a cheap model costs a fraction of what a larger, more complex request on a premium model does. Concurrency, meaning how many agent runs you can have going at once, is capped by plan: 1 on Free, 3 on Personal, 10 on Pro, 50 on Enterprise, per Netlify's own published limits.

In practice, for the kind of small, well-scoped edit this page is about, a handful of changes a month fits comfortably inside the Personal or Pro credit pool alongside your normal hosting usage. Where the cost adds up is a vague, open-ended prompt that makes the agent explore the codebase extensively or retry a failing build several times before it gives up, which is exactly why starting with a precise instruction on a low-stakes page matters more than the headline credit price.

Recent model updates worth knowing about

Netlify has been adding new AI models to Agent Runners and its underlying AI Gateway at a fast clip this October: OpenAI's GPT-6.1 Terra landed October 26, Google's Gemini 4 Argon on October 13, Mistral Large 4 (with a very large 524K-token context window aimed at multi-step coding tasks) also on the 13th, and MiniMax M3.1 Flash on October 23, all confirmed on Netlify's own changelog. Netlify also switched new Agent Runners projects to TanStack Start 1.0 as the default framework as of October 13, unless you specify something else. None of this changes how you use the feature day to day, but it's worth knowing the underlying models are actively being upgraded, which tends to mean steadily better results on the same kind of prompt over time.

Three real prompts that work well, and one that doesn't

The difference between a clean result and a frustrating back-and-forth usually comes down to how specific the instruction is. "Update the footer phone number from 555-0100 to 555-0142 on every page" works well because it names the exact old value, the exact new value, and the scope. "Add a new page at /service-areas listing Downtown, Riverside and the Northside, each with a short paragraph, linked from the main navigation" works for the same reason: it names the URL, the content, and where it should be linked from. "Make the homepage look more modern" is the one that doesn't work, not because the agent can't try, but because "modern" means something different to every person who types it, and you'll likely spend more credits refining a vague result than you would have spent being specific the first time.

A practical habit worth adopting: write your instruction in a notes app first, read it back as if you were the agent with zero context on your business, and only then paste it in. Most of the vague-prompt problem disappears once you catch yourself writing something that assumes the reader already knows what you mean.

What I'd actually tell a small business owner

For the genuinely small stuff, swapping text, updating a price, adding a simple new page, this is close to the best version of "AI replaces a $100 invoice" that currently exists, and I'd use it without much hesitation. Where I'd pump the brakes is anything where a mistake is expensive to notice late: pricing tables, legal or compliance text, anything feeding into a form that writes to a database. Not because the AI can't technically make the edit, but because nobody on your team is likely to catch a subtle error in those specific spots the way they would a visibly wrong headline.

The other thing worth being honest about: this tool is genuinely different from hiring a freelancer on retainer or using a competing AI builder like Lovable or Bolt. Those tools and people bring judgment to an ambiguous request. Agent Runners does exactly what you ask, which means the quality of your result depends heavily on the quality of your instruction. If you find yourself needing to explain nuance, business context, or "it depends" reasoning to get the right outcome, that's usually the signal you've crossed from a prompt into a conversation, and a person is still the better tool for that.

When to skip this and call a developer

Skip the AI route entirely for anything touching payments or customer data, full stop, and get a developer to review the change instead. The underlying risk isn't that the agent will obviously break something, it's that it can produce code that looks fine and runs without errors while still handling an edge case wrong, and that kind of bug is exactly the type you don't want anywhere near money or personal information. If you're on a site with split testing enabled, Netlify's own documentation flags that Agent Runners shouldn't be used alongside it at all, so check that first if you're not sure.

And if the simple options genuinely stop working for what you need, whether that's Agent Runners hitting its limits or a freelancer's retainer not covering the complexity of the request, that's the point where a proper custom build, done by someone who can own the whole picture rather than respond to one prompt at a time, usually pays for itself. I build exactly that kind of project for clients who've outgrown the prompt-and-preview loop, or who have a half-finished AI-built site that needs a developer to actually finish properly.

How I'd approach this

If you're already on Netlify and the edits you need are genuinely small and well-defined, I'd start with Agent Runners directly: set a credit limit, try it on one real but low-stakes change, and judge the result on your own site rather than a demo video. If the pattern of requests you're making starts looking less like "change this text" and more like "build this feature," that's usually my cue that a scoped custom build, with a fixed price agreed up front, is the better next step rather than stacking prompt after prompt on a tool built for small, discrete edits.