Tyren Temple
Tyren Temple, photographed in profile at golden hour
Tuned for

Design and build.Same hands.

Product design & AI  |  Fintech, GreenTechWellington → Sydney
Designer to designer

I build the systems that make AI reliable.

The work I value most sits between product and engineering, closing the gap where design, code, and systems meet. Same instinct drives the tabletop games I design on the side.

  • I design and ship my own front-end PRs, in one loop.
  • I treat design systems as products, judged by whether a non-designer can build with them.
  • I make them machine-readable for AI agents, and stay honest about where the tools fall short.

Selected work

2026

Design-system infrastructure · Brighte

AI agents kept guessing at our UI. I gave them a source of truth.

Design systemsAI
Challenge
The component library worked for humans, but an AI coding agent had no reliable source of truth to read. Point one at the system and the output was inconsistent, throwaway code that needed correcting everywhere.
Solution
I built a context and semantic layer on top of the system: every component specified with the tokens it uses, how it composes, and what an agent must never generate, with pattern rules above describing how components assemble into whole surfaces. Agents read the layers in sequence instead of inventing.
Role
Product Designer and design-system custodian at Brighte. My idea, and I led it, working hand in hand with engineering.
Impact
The same brief run through independent agents in fresh sessions now lands consistent structure, components and tokens: close to draft feature code where there was throwaway output before, with far less rework.
Coming soon
Coming soon
Coming soon
Coming soon
2026

AI-powered vetting platform · Brighte

Vetting installers took a month. Now half a day.

Product designAgentic workflow
Challenge
Vetting took up to a month across four systems, with 58 minutes of manual work per application and a backlog building. Worse: more than half of accredited vendors never went on to write a single deal.
Solution
I reframed the brief from "redesign the form" into designing the data layer for an AI system. Vendors move through a structured intake that hands the agent clean data to act on, and the admin portal surfaces the workflow's results so humans keep the judgment calls.
Role
Product Designer at Brighte. Reframed the brief and owned every vendor and admin surface end to end, in a pod working hand in hand with engineering.
Impact
Approval to accredited went from 17.5 days to half a day, and decisioning from 58 minutes of manual work to under 4. 89% of vendors who started the form finished it.
2023

Impact dashboard · Trees That Count

One dashboard, two users who wanted opposite things

Product designDual-sidedData viz
Challenge
One dashboard serving two groups with opposite needs. Planters needed project tools and compliance; funders needed impact they could report on, and were requesting manual reports just to get it.
Solution
Same data underneath, so I refused to fork the product. One source of truth, surfaced differently for each side, and shipped incrementally so value landed early instead of in one big bang.
Role
Product Designer at Trees That Count. Led the redesign end to end: research, strategy, design, delivery.
Impact
Planter task completion went from 73% to 96%, funder engagement lifted 115%, and average session time tripled from 30 to 90 seconds.
2025

Design system · AddressFinder

Figma said one thing. The code said another.

Design systemsTokens
Challenge
The portal redesign was stuck because design didn't match the build. Designers and developers were describing the same components in different languages, with no shared source of truth to settle it.
Solution
I audited how their site was actually built in Tailwind and built the system on top of that reality: 240+ tokens and 85+ components named to the code's conventions, so the thing in Figma and the thing in production are finally the same thing.
Role
Product Designer (freelance). Owned the audit, the token architecture, the component library and the handover.
Impact
The team designed 30-40% faster, the Head of Product's estimate from daily use, and a non-designer marketer shipped on-brand pages without waiting on design.
2026

Card game · Personal project

Financial lessons don't stick, so I made losing the lesson

Game designSystemsPrint
Challenge
Financial literacy gets taught and doesn't land: around 70% of NZ schools teach it, but only 19% of students feel they learn anything. It arrives as homework, not as something relevant to your actual life.
Solution
A card game you learn by losing. Take a risk, occasionally get burned, and the consequence does the teaching, no lecturing required. Balanced across 12-plus rounds of playtesting with 50-plus people.
Role
Solo designer. Concept, mechanics, balance and production, publishing in 2026.
Impact
Over 90% of players who started high-risk shifted to cautious play after a few games, and players could explain dividends, interest and debt in their own words afterwards.

What I'm known for

Context engineering

Systems built for machines to read

The decision I'm proudest of: stop treating AI's inconsistent output as the problem and treat it as missing context. Every component spec'd with what an agent must never generate.

Opinion

AI is enablement, not deskilling

AI lets skilled people do things that were too hard before. The risk is real, but it sits with the person, not the tool.

How I work with it

A multi-agent stack, weighted to my gaps

I deliberately weight my stack toward technical agents because my engineering background is lighter, and I separate the thinking tool from the building tool.

Honest about it

I challenge what the AI hands me

These tools favour their own output and will happily agree with you. The moment you stop challenging that is the moment it quietly steers you.

Designer to designer

You've read the whole thing. Now ask the hard ones.

Craft, AI, design systems, the seam with engineering. Ask the brain anything, it answers in Tyren’s own words.

Ask anything, or paste a job description for an honest fit read. Pasted JDs aren't stored.