Agent Console
Operations console for AI agent runsThat I understand the shape of agent products — and which single view carries the weight.
Killer detail
Replayable execution trace
Stack
Next.js · React 19
Built in
~2 days
Design Engineer · Kuala Lumpur, GMT+8
One person, from the first sketch to deployed code — usually two to three weeks. I work on interfaces where the domain is genuinely hard: trading, risk, agent tooling. Anything where the data is dense and being wrong is expensive.
Twelve years in product design. The last four in fintech and AI tooling.
What I do
$12k – $35k
2 – 4 weeks
A real surface of your product, designed and shipped. Not a Figma file for someone else to interpret — a running application your team can pull and extend.
Best when: You have engineers but no one who owns how it looks and feels.
$10k – $25k
2 – 3 weeks
A token layer, a component library, and a documentation site — plus one of your real products refactored onto it, because a system nothing runs on is a set of guesses.
Best when: Your interface drifts a little further apart with every sprint.
$100 – $200 / hr
Ongoing
Design and front-end capacity on your team for a stretch. Useful before a launch, during a rebuild, or when a founder is still doing the design and would rather not be.
Best when: You need the capacity now and the scope is still moving.
Fixed price against a written scope, agreed before any money moves. Anything that changes the date or the price gets an amendment in writing first.
Work
All four are self-initiated. Client work is under NDA, and I would rather show you something I can talk about honestly — including the parts that went wrong.
That I understand the shape of agent products — and which single view carries the weight.
Killer detail
Replayable execution trace
Stack
Next.js · React 19
Built in
~2 days
That dense financial data can stay legible, and that the keyboard can do everything the mouse does.
Killer detail
Brushing re-runs every panel
Rows
2,000 virtualised
Built in
~1 day
That I build the layer other people build on — and that I make a real product consume it first.
Killer detail
Two independent axes
Tokens
64 primitive · 37 semantic
Built in
~2 days
That something can look expensive without being expensive, and that performance claims are measured.
Killer detail
120fps, 0 asset bytes
Draw calls
3
Built in
~half a day
How it works
01
Forty-five minutes on what the thing actually has to do. You get a written scope with a fixed price and a date — before any money moves. If the scope is wrong for you, that costs you forty-five minutes.
02
You see work in progress two or three times a week, in a link you can click rather than a deck. Feedback lands while it is still cheap to act on. No two-week silence ending in a reveal.
03
Because the same person does both, there is no handover to lose things in, and no design that turns out to be impractical three weeks after it was approved. What you review is running code.
04
The repository, the design system, and a written record of why each decision went the way it did. Your team should be able to extend the work without needing me — that is the point.
I am in Kuala Lumpur. My evening is a European morning, so there is real overlap with the UK and the EU. For US teams the overlap is thin — which means I work the way remote work is supposed to work: written updates, a clickable build every couple of days, and meetings only when a decision genuinely needs one. Most weeks that is zero.
Who you would be working with
I have spent twelve years designing product interfaces, most recently in fintech and AI tooling. That means I have read enough trade blotters, risk views and execution traces to know what a practitioner checks first — and it is never the layout.
I work as one person on purpose. Design and implementation in the same pair of hands removes the handover, which is where most of the cost and nearly all of the compromise usually lives.
I use it heavily. It is the reason a two-person job takes one person two weeks.
It is also why the market is now full of people who can generate a competent-looking interface in an afternoon. So the honest question is not whether I use it — it is what I add on top.
What I add is the review. Knowing that a design system with no product running on it is a set of guesses. That a correlation matrix computed once is quietly lying to you. That a grid which drops keypresses under key repeat will be found by a trader on day one and never trusted again. None of those are visible in a screenshot, and none of them get caught by the generating.
Twelve years is what makes the review worth paying for. The AI is a faster pencil.
Straight answers
No. All four are self-initiated concepts, built to show how I work rather than to dress up somebody else’s product. Client work is under NDA and I would rather show you something I can talk about honestly, including the parts that went wrong.
A repository you own, deployed and running, plus the design system it is built on and a written record of the decisions. Not a Figma file with a handover meeting attached.
That is the usual arrangement. I take a surface end to end and stay inside your conventions — your stack, your review process, your component library if you have one worth keeping.
It usually does. Small changes get absorbed. Anything that moves the date or the price gets a written amendment before I start on it, so there is no invoice at the end that you did not see coming.
Rarely, and only when it genuinely fits. The value here is the two together — separating them puts back the handover I am charging you to remove.
Next step
Forty-five minutes, no deck. If I am not the right person for it I will say so on the call, and I will usually know someone who is.