Curiosity, experiments, and the work in between
I work in fintech sales and business development. I also like finding out whether an idea can be made to work, which is how a question sometimes turns into a tool, a diagram, or this website.
I talk things through, try a version and ask what I’ve missed. AI gives me more room to do that. I don’t necessarily finish sooner; sometimes I just get further into the question.
A few things I’ve been working on ↓Some of the work behind the work
The useful part is the example, including what still needs work.
01 / A project I built
I built a way to describe a payment and explore the steps as a diagram. It gives me something concrete to work through: who is involved, what moves, and what the explanation leaves out.
The check: a convincing diagram still needs to match the payment flow. Drawing a line between two boxes doesn’t establish that money moves that way.
Explore Flow of Funds (opens in a new tab)
02 / Live, and still being edited
AI helped with the code, image editing and reviews. I supplied the experience and kept correcting the story. The homepage portrait is AI-assisted, and its setting is composed.
I helped grow the client base from one to twelve.
That word keeps my part in the result clear. The same review needs to catch a page that sounds impressive but no longer sounds like me.
See the work and the context ↗The Toronto detail started with the time and directions. When I considered adding weather, I wanted it to do something more useful than display a temperature: help someone visiting decide what to bring.
That became “Come say hello,” with a little advice about layers, a weather-dependent joke and a way to arrange a conversation. The observation source and time are there if you want to check them. If the data is stale, the advice disappears.
It’s a small design decision. Whether visitors find it useful is something I’m still learning.
03 / A small public experiment
The small markers on my watch, shirt, shoes and bike open details and retailer links. I’m exploring what happens between that moment of interest and a real retailer cart.
The check: the right item, available option, current price and an honest handoff to the seller. A clickable marker is only the beginning.
The public version uses retailer links. A separate prototype has reached a retailer checkout in testing; live inventory and agent checkout are still being developed. Purchases finish with the retailer.
So I don’t start from scratch every time
I use Basic Memory Cloud for selected background and working preferences. The aim is to keep the useful context current, with the actual work staying in its own files and tools.
How I write, the work I do, and the purpose of a project. A small set of maintained notes.
Basic Memory CloudFind the point, keep my voice, challenge the claim. Skills give an assistant a specific job.
Written instructionsThe goal, what’s decided, what’s open and the next move. Enough to pick the work up again.
The current projectThe intention is to reuse the relevant context across tools that can access it. Each connection needs checking. A shared note doesn’t make separate assistants share a conversation, and saved context can still be wrong or out of date.
This page shows the approach and public examples. Private notes, conversations and account configuration stay out of the public material.
A setup that changes as I use it
My daily drivers, because I’m basic. Useful for another version, another angle, or a question I haven’t asked yet.
Code, tools and agent workflows. For projects such as Flow of Funds and Polaris, I also use lower-cost APIs where they fit the job.
Dictation helps me start. The notes and instructions help an assistant work with the idea, including the corrections.
Things I’ve experimented with alongside my usual tools. Trying something doesn’t automatically make it part of my everyday workflow.
This is a snapshot of my setup, not a ranking or a recommendation to subscribe to everything on the list.
Something you can take away
These three are public adaptations of instructions I use. Change them to fit your own work.
Work out what a wandering conversation is getting at, without treating every idea as a decision.
It can make uncertainty sound settled. Ask for an interpretation you can correct.
Make the writing clearer while keeping the detail, uncertainty and humour that belong there.
Banned words won’t create a voice. Use real writing examples and read the result yourself.
Find the unsupported claim, missing context or misleading impression before the work goes out.
A second assistant can agree too easily or invent objections. Ask for something specific you can verify.
These were written for the site and aren’t established parts of my routine: Rehearse a decision and Design a model test. Both can help frame a check; neither substitutes for evidence from actual people or use.
Download decision exercise ↓Download model-test exercise ↓Each ZIP contains a plain-text SKILL.md file. Read it before using it. You can also take a passage and use it in a conversation.
If you want to talk about the work, the experiments, or something you’re building:
Get in touch ↗