The setup: Customer Zero for Agent OS

Over the course of this year, the people who actually run contentstack.com—content managers, marketers and the Agent OS product team—have sat down together to test Agent OS and Polaris, using screen shares of what worked, what broke and what needs to change before it ships.

We call the program Customer Zero. The idea is simple: Before we ask a customer to trust Agent OS with their content operations, we run it through ours first.

What we got wrong in the beginning

The initial sessions were about finding the edges.

• Agents live in production, no safety net. Our early sessions surfaced custom agents rewriting headlines, translating live entries and running publish-triggered checks directly in our production environment, with no staging or preview step to catch a bad output before it went live. The team paused every running agent that day and moved all content-editing and content-creation testing to a dedicated test stack. That rule has held ever since.
• Onboarding assumptions, exposed. New users hit confusing account-naming conventions, unclear management-versus-delivery tokens and agent templates that failed without explanation. Several sessions were spent just mapping where the product needed better in-context guidance.
• Defining what Polaris is actually for. Early tests asked Polaris to do things it was never built for, like diagnosing a front-end rendering bug or scanning an entire domain for broken links. The team used those failures to draw a clear line: Polaris owns CMS-side work for content editors and website managers, not code or presentation-layer debugging.

All of this feedback went straight back to the product team building Agent OS, which is the point of Customer Zero: Friction found here gets fixed before a customer ever hits it.

What started working

Sometime later, the sessions had shifted from “what’s broken” to “what can we ship.” A few examples, pulled straight from the test stack:

• Glossary content, minutes instead of days. Using Polaris, one team member drafted and published five glossary entries in about 20 minutes. Built manually, field by field, the same work would have taken days.
• FAQs written in the brand’s own voice, not a freelancer’s. Polaris pulled from Contentstack’s own brand kit and voice profile to draft FAQ sections for blog posts, then published directly to the live entry after a quick human review. No outside copywriter needed.
• A recurring SEO weak spot, fixed in bulk. Contentstack’s generative engine optimization reports kept flagging weak cross-linking between related articles. Polaris tagged AI-related blog posts and added cross-links between them automatically. That used to require someone combing through the CMS by hand.
• Bulk Semrush and metadata fixes, one pass. The team used Polaris to find and fix duplicate meta titles and Semrush-flagged errors across several entries at once, then fill empty Open Graph fields the same way.

The common thread: The team stopped chasing ambitious, unproven use cases and started with the repetitive work that was already eating hours every week. As one team member put it in a session, “let’s knock out the low-hanging fruit that can save us time immediately” before testing whether AI-driven content changes actually move AI visibility scores.

How this supports GTM

This is the part that matters beyond our own content team. Agent OS is great for cleaning up blog posts, but more importantly, it’s producing the assets behind live GTM campaigns.

Our account-based marketing motion needs voice profiles tuned to specific target accounts and campaign angles. Building those by hand, one account at a time, is exactly the kind of repetitive, brand-constrained writing task Polaris is suited for. Using the brand kit and voice-profile tools tested in Customer Zero, the marketing team is now building unique voice profiles directly in Polaris from existing campaign copy, rather than starting from scratch for every audience.

It’s a small example with a bigger implication: The same agentic platform we’re selling to enterprise marketing teams is already load-bearing inside our own demand-generation engine. That’s also why the team is planning to demo these use cases internally at our monthly AI town hall, so sales and other departments can copy the same patterns instead of rediscovering them independently.

What Customer Zero changed on the roadmap

In addition to improving how our team uses Agent OS, feedback from these sessions directly shaped the launch of the following platform enhancements:

• Skills, a feature letting users save a prompt as a reusable, named command, came directly out of the team asking for a way to stop re-typing the same instructions every session.
• Bulk-operation summaries, arriving by the end of August 2026, respond to a specific complaint: Approving every single change in a large batch job, one click at a time, was safe but slow.
• Deeper GEO integration is moving forward after the team found itself manually screenshotting GEO recommendations into Polaris as a workaround, because a future GEO product we’re working on wasn’t quite in sync with Polaris yet.

That’s the actual value of running Customer Zero: A feedback loop with a paper trail from a real complaint to a shipped fix.

Why we’re telling you this

Most vendors will tell you their AI product is ready for your content team. We’d rather show you the time it took ours to trust it with real work, including the week it wasn’t ready. That’s the whole premise of Contentstack on Contentstack: We don’t publish claims about our platform that we haven’t tested on our own site first.

Agent OS and Polaris now write marketer-approved FAQs, clean up SEO debt and build the campaigns you’re reading this from. If you want to see what that looks like for your team, talk to us.