“So, how are you using AI?”

This is the question of the last 12 months. I’m asking it, seeking the answer to it or I’m answering it.

It’s like we were all using abacuses (abaci?) and now calculators are available. Everyone wants to know how best to use this new thing.

Like most folks in product or technology I started where you would expect: ChatGPT, Perplexity replacing Google. I was doing things like asking for help with SQL queries so I didn’t have to bother engineers. It started as a time saver.

But the real shift happened when I stopped treating AI as a question-answer tool and started building systems that work for me persistently.

From prompting to building

Late 2025 – Creating a product sparring partner

I created a “Product Coach” using a Custom GPT in ChatGPT. I gave it my current work context and the perspectives of some product thinkers I follow: Teresa Torres, Dan Olsen, Jeff Patton, Matt Le May.

Using default prompts: “How should I prioritise my day?” this gave me a sparring partner to catch things I might not have thought about when drafting product requirements docs, roadmap thoughts etc. I later converted it to a skill that could be used with different AI harnesses like Claude Code.

Early 2026 – Building a second brain

Inspired by Dave Killeen’s Dex project, I started using Claude Code heavily and built my own “Second Brain” called Mind Palace.

Rather than having one-off prompt sessions, this was an entire change to my workflow. I can now easily store and update context about current work projects, meeting notes, as well as build and call skills (sets of instructions about how to approach and complete repeatable tasks).

The whole system compounds with context and learning over time. I can prompt: “Should we save anything before I close this session?” and the system gets smarter. No more starting from scratch.

March 2026 – Sharing the capability for org-level benefits

The start of 2026 felt like a mad scramble for a lot of us to experiment with AI personally. But for me it’s important these capabilities get shared. Then, more people can get the benefit of AI in their day-to-day and we’re not all recreating the same thing.

So in March I created the internal-skills repository at Validation Cloud to help us internally share useful AI capabilities, seeding it with a first set of skills I built:

• rpc-check to be able to make JSON RPC calls against Validation Cloud nodes and others
  
• vc-data-analyst allowing anyone on our team to ask natural language questions and get usage data about our customers in short summaries and with generated charts showing the most important trends (no SQL required)
  
• vc-design-agent a design partner with knowledge of our branding guidelines, design system and component library to help create mockups and review additions to our products
  
• vc-release-notes Weekly release notes generator looking at a number of our internal repositories and producing a pull request on our docs repo for manual review before publishing

Others in the team have already built on top of these and introduced their own skills which is great to see.

Late March 2026 – Shipping and getting usage on my own project

I built and shipped Tailwise, a wind-optimised cycling route recommender for my local area.

I’d used AI for personal projects before but this was a real 0->1. I created something I can use every week along with people at my local cycling club. There’s been some good feedback and reaction so far from my cycling club, here’s a review I enjoyed:

“Just had a go and it works really well. It brings up lots of routes and cafe choices. I especially like that one can select when they want to ride, then it says which advantage of the route based on wind direction. I think you just created a more useful app than Komoot.” – James, local cycling club member

Wrapping up

Initially in early 2026, using Claude Code and Codex in the terminal felt strange to me.

Now, I find myself opening Mind Palace multiple times a day from the terminal. I have different concurrent sessions running at any one time, covering: drafting documents, pulling usage data/trends, discussing things with product coach, building mockups or proposing changes to our product front-end.

Invoking skills for tasks and having /codex review any code generated by Claude Code is becoming a regular workflow.

In the space of a short few months: I’ve found myself progressing from trying AI tools, to building my own systems, to sharing those skills/systems so a whole team of people benefit. I believe that last step is where a lot of the leverage sits.

What’s next?

Well, the space is changing a mile a minute and it’s hard to keep up! Some of the next things I want to explore:

• Running things on a schedule where it makes sense – e.g. my initial release notes skill now runs on a schedule after collaborating on it with a team mate (shout out Willem!)
• Trying to keep up to date with changes to AI harnesses & models to get more leverage in my day-to-day as well as at the organisational level
  

Resources I’ve found helpful and why

• Dave Killeen talking about his Dex project with Ed Biden of Hustle Badger.
  • How it helped me: Dave inspired me to try the Dex project, make a contribution and ultimately build my own system in Mind Palace which I now use all the time
• Sim Superville’s PM Guide to Claude Code
  • Sim’s overview helped me get comfortable with the basics of Claude Code in the Terminal and other options I could use. My preferred mac client is now Ghostty as recommended by one of our engineers.
• Claire Vo’s How I AI Podcast – On YouTube & Spotify
  • Being a visual learner, it helps me to see things and how people are actually using AI is what Claire’s podcast goes deep on, lots of examples from different folks.
• AI News & Strategy Daily by Nate B Jones
  • A lot of thought provoking content on the latest developments in AI models, tools and everything else AI related.