Rebuilding a Website for the Agent Era: From CMS to Agent‑Native Content Operations

The Problem: Bifurcated Context

When Prefect’s marketing site ran on a conventional headless CMS, the content lived in a remote store that the AI agents could not directly access. The agents could read the code but had no visibility into the rendered content, forcing human intermediaries to bridge the gap for every change. This “hidden tax” of the CMS abstraction turned a seemingly simple update into a multi‑step, human‑driven process.

“When you’re trying to work with AI coding agents, a CMS creates a wall between the agent and the content it needs to modify.” – Prefect VP of Product

The Approach: Multi‑Model Orchestration

Instead of a straight migration, the team treated the rebuild as a greenfield project that could pull from the existing CMS. The key was matching the right model to the right task:

The planning phase involved Claude reflecting on the work needed and even drafting job descriptions for hypothetical hires. This meta‑planning fed directly into the sub‑agents that executed the tasks.

Eating Our Own Dog Food

Prefect’s open‑source workflow orchestration platform and its commercial product provided a unique advantage: the team could hook the agent to Prefect’s own documentation MCP server. This automated quality‑assurance step flagged inconsistencies between marketing copy and the actual product documentation—something that would have been tedious to catch manually.

The Migration Was Actually the Easy Part

Migrating the blog and static pages to MDX was straightforward: a read‑only CMS token let the agent pull data, convert it to Markdown, and generate open‑graph images. The real work lay in redesigning the entire site to be agent‑native, enabling CI/CD for content and automated style‑guide feedback.

Tokens used: ~25M (mostly Claude)
Commits: 290
Estimated effort: 6 weeks → Actual: 1 week of focused work

The Real Unlock: Agent‑Native Content Operations

With all code and content in a single repository, the team could:

• Run CI/CD pipelines that include content quality checks.
• Provide automated feedback to writers before human review.
• Use Linear and Devin to triage and implement changes autonomously.

In the first week after launch, a competitor’s LinkedIn post about Airflow 2’s end‑of‑life was turned into a fully fledged landing page in roughly 30 minutes—half the time it would have taken in the old workflow.

What This Means

Lee Robinson’s observation that “the cost of abstractions with AI is very high” rings true. Infrastructure designed for human editors—CMS abstractions, separate code and content stores—now becomes a bottleneck for AI agents. The solution is to tear down those walls and build highways that let agents move freely.

“If you’re building tools yourself, use them to validate your own claims. Hook up your docs MCP server and let the agent catch drift.” – Prefect VP of Product

By making the website agent‑native, Prefect has demonstrated that a well‑orchestrated AI stack can compress weeks of work into minutes, setting a new benchmark for content operations in the age of generative AI.

Source: Prefect Blog – Rebuilding Our Website for the Agent Era