I build the loop that keeps your docs accurate, release after release.

The problem

Most technical content goes stale the day it’s published, and AI agents act on it without questioning.

The method

I build the Knowledge Loop: an improvement loop orchestrated by AI agents that collects signals (GitHub, Stack Overflow, Slack, forums, chats, emails), separates real problems from noise, and turns them into fixes in the docs and the product. The result is a single source of content, kept current with the latest product releases.

Proof - turning scattered customer feedback into weekly fixes

Whether your team is part of a large enterprise or a fast-moving startup, this method holds up. I’ve applied it in both, defining the content, prompt and context engineering strategy: modular content, prompts that guide language models, and the right context for AI agents to provide secure and accurate recommendations to end users.

The situation

Customer feedback reached the product and content teams through many channels: Microsoft Q&A, the Azure feedback portal where product teams prioritize features by upvotes, Stack Overflow, Reddit, LinkedIn and other internal forums. No single team triaged it all. Each team received a constant stream of requests, much of it noise or duplicates, and very little of it turned into actual improvements in the docs or the product.

What I built

I designed and built a feedback triage agent, then integrated it into Microsoft’s internal AI portfolio.

• It connected to every feedback channel (public and internal) and grouped similar requests across channels.
• Every week, it produced a report that prioritized what needed to change, either in the docs or in the product.
• For each item, it drafted the actual change. The human only had to approve or improve it.

The Knowledge Loop Build

If the assessment shows the loop is worth building, and you’d like my help, here’s how I’d build it with your team. The report and the plan from the assessment are yours either way.

First months — build:

• Content: restructure your content into modular, agent-readable knowledge (docs-as-code)
• Context: define what knowledge each agent receives, when, and in what format (Markdown, llms.txt, skills, search)
• Prompts: write and test the instructions that guide the agents, both those that answer users and those that run triage
• Connect your feedback signals into a single pipeline
• Set up agent triage with human approval, routing fixes to the docs or the product
• Add automated checks that keep the docs and the agents’ answers accurate (docs-as-tests)

After the build — care, for as long as it’s useful to you:

• Run and tune the loop together with your team
• Monthly summary of gaps found and fixed
• Train your team to own the loop