The AI Support Agent That Answers About Features Before They Ship

Most teams rolling out AI support hit the same wall: the assistant is only as good as the content behind it. Point it at a stale knowledge base and it will confidently give users the wrong answer — which is exactly how you train people to skip the AI and ask for a human instead.

The fix isn't a smarter model. It's a knowledge base that keeps itself current. Trond-Daniel Kvalvik, Product Manager & Team Leader at Mystore — Norway's largest e-commerce platform, part of the Visma Group, serving more than 1,100 online stores — built one that does. Here's how his setup works, and how to build your own self-updating knowledge base on Product Fruits.

Why AI support gives wrong answers (it's rarely the model)

When an AI support agent answers incorrectly, everyone blames the model. Usually the problem is more mundane: the content it's reading has gone out of date. Most modern support assistants work by retrieval — they search your knowledge base, grab the most relevant article, and answer from it. If that article is stale, the answer is stale. There's a name for this: knowledge base rot.

For teams that ship quickly, rot sets in fast. Every release nudges the existing docs a little further from reality, and because articles are usually written by hand days or weeks later, the assistant is permanently a step behind the product. Users ask about the newest thing, get a vague answer, and stop asking.

Trond-Daniel sees this as the real adoption blocker:

“Many customers don't want to talk to AI. They want a person… So on those things, it's really important to keep your sources up to date.”

Keep the sources fresh and the same model suddenly looks a lot smarter.

The fix: generate your docs from code, then sync them by API

Instead of writing articles by hand, Mystore generates them. Trond-Daniel built a workflow where Claude writes knowledge base articles directly from their source code, then pushes them into Product Fruits programmatically:

“I have a workflow with Claude that creates knowledge base articles from code, and then pushes it all into Product Fruits using the Knowledge Base API… when we launch something new, Elvin already knows about it. So he can answer questions even before the product has actually launched.”

Because the docs come from the same code that defines the feature, the knowledge base is ready before the feature ships. Elvin, Product Fruits' AI assistant, can answer about a feature on day zero, not two weeks after launch.

What makes this durable

It removes the bottleneck without removing people. Trond-Daniel is clear that AI is “a tool to speed up productivity, not necessarily to replace anyone.” Mystore steers users to Elvin first because it's the only way to learn where the content is weak. They committed fully once a custom API trigger let them retire the direct-to-human button: every conversation now starts with Elvin, which drafts a structured escalation email on the user's behalf when it genuinely can't help.

It produces numbers you can point to. Within 90 days, Elvin autonomously resolved close to 60% of support inquiries, and Mystore's support team says it absorbs the minor issues so people can handle what actually needs a person.

Curious what an AI assistant could resolve on your own product? See it on your use cases — book a demo.

How to build a self-updating knowledge base on Product Fruits

The pattern has four parts, and all of it uses features available today.

1. Generate or maintain your articles programmatically. Mystore generates them from code with Claude, but the source can be anything — existing docs, release notes, or an LLM workflow. What matters is that the content is produced programmatically, so it can be regenerated whenever the product changes.

2. Push them with the Knowledge Base API. Product Fruits exposes a REST Knowledge Base API at api.productfruits.com/v1/knowledgebase. The import endpoint — POST /v1/knowledgebase/import — creates or updates articles in a single call, keyed by a correlationId you control, with full multilingual support. Because it's idempotent on that ID, you can re-run it on every release: new features create new articles, changed features update existing ones, and you never end up with duplicates.

3. Point Elvin at the knowledge base. In Elvin's settings, add your Product Fruits knowledge base as a source. Elvin can draw from the built-in knowledge base, external URLs and sitemaps, and custom text — so the freshly imported articles become answerable content automatically.

4. Close the loop. Review the conversations Elvin couldn't resolve, find the gaps, and improve the source. Trond-Daniel's advice is to switch it on early and learn from real questions rather than waiting for “perfect” docs:

“You can write beautiful sources for Elvin, but without testing… it's very hard to find out where you need to improve it.”

The takeaway

AI support lives or dies on content freshness. Mystore's answer — generate docs from code, sync them through the Knowledge Base API, let Elvin serve them — turns a SaaS knowledge base that’s always slightly behind into one that's ready before launch. The payoff is an AI agent customers actually trust, and a self-resolve rate near 60% inside three months.

If your AI support is giving shaky answers, start with the content, not the model.

Want to wire up your own pipeline? Explore the Knowledge Base API, see how Product Fruits handles AI support, or book a demo and we'll walk through it.

FAQ

What is a self-updating knowledge base?

A knowledge base whose articles are produced and refreshed programmatically — from source code, release notes, or an automated workflow — rather than written by hand. Because the content regenerates whenever the product changes, the AI assistant reading it stays accurate.

Why do AI chatbots give wrong answers?

Most often because the underlying documentation is out of date. Retrieval-based assistants answer from whatever article they find; if that article is stale, so is the answer. Fixing the content usually fixes the answers.

How do I keep an AI knowledge base up to date?

Automate it: generate articles from a reliable source (like your code), push them into your knowledge base via API on every release, point your AI assistant at that knowledge base, and review unresolved conversations to close the gaps.

Can I generate knowledge base articles from code?

Yes. Mystore uses Claude to draft articles from source code and imports them with the Product Fruits Knowledge Base API (POST /v1/knowledgebase/import), which creates or updates articles idempotently.