AWS cuts marketing page production from hours to minutes with agentic AI

AWS has described how its marketing team moved the assembly of content pages to a semi-automated workflow with agentic AI. The solution, built on Amazon Bedrock together with Gradial, cut the preparation of a single web page from about four hours to ten minutes without giving up quality checks.

Where the process stalled

AWS describes a typical marketer's workflow like this: a campaign brief, a call with the digital team, a place in the backlog, then manual page assembly in the CMS. Even when the task itself is simple, time disappears into block configuration, content placement, approvals and repeated edits. The problem is especially visible in a corporate environment, where a single page must simultaneously satisfy brand rules, SEO requirements, accessibility and internal publishing procedures before going to production on global digital properties.

• Manual assembly of blocks and layouts in the CMS
• Delays from checking text, links and creatives
• Dependence on engineers when ready-made components are missing
• Late checks for SEO, accessibility and brand standards

AWS's key conclusion is that this is not a set of separate small problems but a single systemic process failure. Quality was controlled too late, and most of the time went not on marketing strategy but on mechanical page assembly. As a result, marketers spent hours on operations that can be formalized: choosing the right components, building the structure, running checks and handing the page to publishing without a long chain of manual steps.

How the solution works

The new system is built around Amazon Bedrock and the Anthropic Claude and Amazon Nova models. The marketer describes the task in plain language: which page to build, which blocks are needed and what the output should be. The Gradial agent then interprets the request, determines the page structure, selects components and generates a configuration that previously required knowledge of the CMS's internal logic. In effect, the authoring interface turns from a set of complex forms into a dialogue in which the system breaks the task into steps itself.

A separate layer in the architecture is an MCP server for real-time quality checks. Instead of waiting for a final review, the system checks content against SEO, accessibility and brand-standard requirements during assembly. If an image, text or page structure fails a check, the problem is visible immediately in the same session. This removes the loop typical of corporate marketing, where a page is almost ready but then goes back for rework because of one faulty element.

The last stage is programmatic submission of the result to the corporate CMS through a proxy layer. It does not replace the publishing system; it connects the agent to the existing infrastructure so the page is created within the familiar data model and governance rules. As a result, AWS did not have to rebuild its whole publishing process from scratch: the agent automates assembly and handoff, while the CMS control environment stays in place with the necessary permissions, logging and approval stages.

What changed afterward

After launch in production, AWS compared metrics before and after the rollout. The most notable metric is page build time: instead of up to four hours of manual work, it fell to about ten minutes, a reduction of more than 95%.

But the time savings are not the only gain. Quality checking became proactive, the interface for the team simpler, and the process more linear. Where there used to be separate stages of configuration, checking and handoff, most actions now happen in a single flow. For marketing teams, this frees up time for tasks that really affect campaign results: positioning, messaging, hypothesis testing and content optimization. AWS states the project's goal directly: to remove mechanical work that does not create value on its own.

If the approach scales to other digital channels, agentic AI could become not just an accelerator for the CMS but a new operating layer for content teams in the enterprise.

What this means

The AWS case shows a more down-to-earth scenario for agentic AI than the usual chat-assistant demos: not generation for its own sake, but automation of a specific bottleneck in a business process. If an agent can understand a task, assemble a page, validate the result and work with an existing CMS, the company gets not an experiment but a measurable tool with a clear ROI and a direct link to the team's operational metrics, launch speed and publishing cost.