Building AI Products Around Real Workflows
Without a coding background, I act as product owner for AI tools built around workflows I know first-hand, from a voice-led video studio to a content studio and the web app my AI team runs on, defining the problem, specifying the solution, testing it and deciding what is good enough to use.
- Milestone: 3 AI products in use or testing: video, content and team operations
- Milestone: 49 automated tests guard the Content Studio (October 2026)
- Milestone: Self-hosted infrastructure with cost controls
The idea
The fastest way to understand what AI can really do in a business is to build with it. Not demos, but tools I depend on every week, with real users (starting with me) and real consequences when they fail.
My role is that of a product owner: define the problem, map the workflow, write the requirements, choose the models, direct AI-assisted development, test the result and decide what is ready.
Three products
Video Studio. A voice-led workflow for short videos. A voice recording is split into timed scenes, storyboarded, previewed and exported as a finished MP4, with the human creator in control of every scene.
Content Studio. A structured editorial workflow: discovering ideas and community discussions, collecting sources, developing an angle, writing, reviewing and keeping every version. It is covered by 49 automated tests.
The AI team web app. The control room for my AI team: one tab per project, each showing the team’s work as a group chat I can join, approvals from a phone, and a live estimate of how much usage each AI subscription has left.
How I work as a non-coder
- Requirements first. Every feature starts as a written problem and acceptance criteria, not a prompt.
- Version control and tests. Changes go through branches and reviews; nothing goes straight to production.
- Cost discipline. Free and low-cost models first, premium models only where quality demands them, and fixed monthly budgets.
- Honest status. Some parts are in daily use, others are still experimental, and I say which is which.
What I learned
A convincing demo is not a dependable tool. The real work starts after the first impressive output: error handling, cost, human review and whether anyone actually wants to use it the next day.
AI becomes valuable when it improves a process you understand, not when it simply produces more output.
That is exactly the skill companies need now: someone who understands the business process well enough to decide what AI should do, what humans must keep, and how to check the result.
