I make software products. Some are AI-native, others I retrofit with AI.
I'm Principal AI Product Manager at VideoMy. We make employee-generated content easy and scalable for enterprise talent teams.
Reach out on LinkedIn, or read me irregular posts on Substack.
VideoMy - I build AI features that cut video production time dramatically. I bake AI into our core features like script writing, pre-recording checks, post-recording enhancements, video editing and derivative content generation.
I'm a key contributor in internal (hobby) projects from @plannededge.
Goast - A social message app, for group conversations that matter: No admins. No drama. No spam. Just vibes. Goast is an AI-native project.
InvoiceGrid AI - An AI-powered invoice management application for Xero organisations. Invoicegrid is an AI-native project.
Vellma - It started as a personal tool to help me communicate effectively in sensitive matters (conflict, legal conversations, emotionally-loaded issues). Vellma is a privacy-focused AI communication assistant for Gmail.
I built MCP servers for my PM Agents with tools for identifying features, functionalities and current implementation specifications. My PM MCPs enable automatic writing and reviewing of issues/tickets in GH, and automated maintenance of roadmap in GH Projects. Check it in action in Goast App - Roadmap.
Also, Xero Dev MCP Server - An MCP server for developers that mocks Xero API and Auth, as well as common CRUD actions. I made it while working on Invoicegrid.
I opened my Codebook - boostraped my project development setup for Claude Code. Used daily for maintaining best dev and devops practices across product developments.
In 2025, prototyping AI features was relatively easy (and fun). Shipping them well was the hard part.
The technical part is often straightforward (and exciting). The harder questions were about creating new UX patterns, handling new failure modes and managing generative output.
Shipping valuable AI features require new tools - e.g. for observability, guardrailing, and fine-tuning, optimising. If you want to make AI-first products, or even just bake LLMs into an existing application, be prepared to build a different kind of product team. It's not just the stack that's new; the SPLF and methodology are too.
In 2026, as even more powerful models emerge, not every opportunity will need an AI feature; not every problem will demand an AI solution. It's the AI PM's mission to find out which ones do.