---
url: 'https://dri.es/website-migration-costs-are-going-to-zero'
title: 'Website migration costs are going to zero'
author:
  name: 'Dries Buytaert'
  url: 'https://dri.es/about'
date: '2026-10-28T15:18:32-04:00'
license: 'https://creativecommons.org/licenses/by/4.0/'
type: blog
summary: 'What happens when AI makes website migrations almost free? Ten predictions for CMS vendors, agencies, Open Source, and buyers choosing their next CMS.'
tags:
  - Drupal
  - Predictions
  - 'Artificial Intelligence'
discussions:
  - { platform: LinkedIn, url: 'https://www.linkedin.com/feed/update/urn:li:activity:7521310827405266944/' }
published: true
featured: false
id: 6351
---

# Website migration costs are going to zero

Within three years, I expect the technical work of moving a website from one platform to another to cost close to nothing.

The existing site gives the agent a reference implementation. It can crawl matching URLs, diff the rendered HTML, compare screenshots, and, given admin access, run the same publishing workflows to check content changes, moderation states, and permissions.

Comparing the two systems under equivalent inputs is called [differential testing](https://en.wikipedia.org/wiki/Differential_testing). Agents work well when they can check themselves against tests. The existing site provides the baseline.

Some of these changes are not new, but cheap migrations could push them _much_ further:

1. **Switching costs stop protecting vendors.** Plenty of software companies keep customers because switching costs are high, not because their product is the best. When a customer can walk into a renewal with a working copy of their site on another platform, the vendor has to justify its price through the value it delivers.
2. **Proprietary vendors will have to sell the way Open Source companies do.** Open Source companies had to build businesses around free code, so they sell hosting, support, and other services. Customers will keep paying for [what has to keep running](https://dri.es/the-software-business-after-code-scarcity): deploying, securing, scaling, and monitoring.
3. **Agents favor CMSs they can try.** Open Source CMSs can be tried directly, while a proprietary CMS behind a sales call may never get evaluated unless its vendor offers self-service trials and programmatic access. Platforms the agent can access, configure, and [verify efficiently](https://dri.es/friction-abstraction-and-verification) gain an advantage.
4. **Vendors will pay for agents to recommend them.** Whoever controls what AI agents suggest controls who gets considered, and a place on the list will be sold, as search engines sold the top of their results. Vendors will pay for it.
5. **The next lock-in is the agent's memory.** An agent that evaluates, migrates, and maintains a site collects that site's requirements, exceptions, evals, and past decisions. If that memory stays inside one AI provider, changing agents means explaining the site all over again. Site owners will need to keep requirements, evals, and decision logs in a form any agent can read.
6. **Technical requirements become evals.** Instead of relying on vendors' answers to technical questions in an RFP, buyers will build an eval suite. They'll run it against agent-built versions of their own site on each candidate platform. The suite could cover editorial approval workflows, translation management, automated accessibility scans, and page speed. It can even keep running after launch to catch regressions.
7. **Vendors compete with contract buyouts.** Cheap migration leaves the contract as one of the few things still tying customers to a vendor. The current vendor will lock them in with multi-year discounts. The next vendor will pay the overlap: credit for what is left on the old contract, or free service until it ends.
8. **Approved-vendor lists become a moat.** When migration takes hours or days, security reviews, compliance, and procurement approvals become the slow part of switching. Customers will favor vendors their organization has already approved. A supplier already on the list could win the business even when another CMS is a better fit. Getting approved needs to become as easy as getting started.
9. **Agencies will recommend CMSs they don't know.** With an agent doing the build, an agency can sell a platform its people are not experts in. AI-generated proposals and working demos will make it harder to distinguish platform expertise from the ability to produce a convincing pitch. Proof of expertise will matter more, to buyers and agencies alike.
10. **A CMS no longer has to be a ten-year bet.** Buyers often choose an established platform because they expect to live with the decision for years. Cheap migration makes a smaller, younger CMS easier to consider: it can win on how well it serves a particular industry today, without convincing buyers it will meet every future need. A tested exit path becomes more persuasive than a five-year roadmap.

When leaving is easy, vendors, agencies, and Open Source projects have to give users a reason to stay. Even if only half of this comes true or meaningfully accelerates, that is a lot. A lot will depend on how quickly people adopt AI and how well they use it.

PS: Follow the discussion on [LinkedIn](https://www.linkedin.com/feed/update/urn:li:activity:7521310827405266944/).