Notes from the work,written to be used.
Short essays on AI models, technology trends, the stacks we build on and the design language we ship. Each one is written for the person who has to make the decision.
16 September 2026
What changes when software is cheap to write
AI coding tools have made producing code dramatically faster. They have not made deciding what to build, checking that it works or running it for years any cheaper — and that shifts where the value is.
2 min read
2 September 2026
Choosing a frontend framework is mostly choosing a team
The technical differences between mainstream frameworks have narrowed. What still differs is who you can hire, how long the code will be supported, and how much of it you actually need.
2 min read
19 August 2026
Designing interfaces for answers that might be wrong
Traditional software is either right or broken. AI features are usually right, sometimes wrong, and always confident-sounding. The interface has to carry the uncertainty the model won't.
2 min read
5 August 2026
Governing AI use without banning it
Staff are already using AI tools, approved or not. A short, practical policy — backed by tools people actually want to use — protects the business better than a prohibition nobody follows.
2 min read
29 July 2026
Legacy modernisation without the big bang
The rewrite that replaces everything at once is the most reliable way to lose two years. Replacing a system one route at a time is slower to describe and much faster to finish.
2 min read
22 July 2026
A design system for teams without a design team
Most organisations that need consistency cannot justify a dedicated design-system group. A small, opinionated kit — and the discipline to keep it small — gets them most of the way.
2 min read
15 July 2026
Write prompts like specifications
The prompts that hold up in production read less like clever incantations and more like a well-written brief for a new colleague. Treat them as code: versioned, reviewed and tested.
2 min read
8 July 2026
Observability for teams of five
You do not need a platform team to know what your software is doing in production. Four signals, one dashboard and a rule about alerts will cover most of what a small team needs.
2 min read
1 July 2026
Passkeys are ready. Your login page probably isn't.
Phishing-resistant sign-in now works across the devices your customers already own. The technology is done; what remains is product design and a sensible migration plan.
2 min read
24 June 2026
Static first: why our own site ships as plain files
No server, no database, no framework runtime — and pages that load before you finish blinking. Static generation is the most underrated architecture for most business websites.
2 min read
17 June 2026
The real cost of an AI feature
The token bill is the smallest line in the budget. Evaluation, review, support and change management are where AI features actually cost money — and where the business case should look first.
3 min read
10 June 2026
Motion with a job to do
Animation is either explaining something or getting in the way. A short guide to the motion that earns its milliseconds, and the kind that should be cut.
2 min read
3 June 2026
Data contracts: stop fixing pipelines downstream
Most broken dashboards were broken upstream, by a change nobody announced. A data contract makes the producer of the data responsible for its shape — and makes the break loud instead of silent.
2 min read
27 May 2026
Two weeks to a decision you can defend
Why we start almost every engagement with a fixed-fee fortnight, and what a good diagnostic leaves behind even if you never hire us again.
2 min read
20 May 2026
Small models, local data
Not every AI task needs the largest model in the world. For privacy-sensitive, high-volume work, a small model you control is often cheaper, faster and easier to defend.
2 min read
13 May 2026
WebGPU and the return of the rich web
Real-time 3D, large visualisations and on-device AI now run in an ordinary browser tab. The question is no longer whether the web can do it, but whether your content deserves it.
2 min read
6 May 2026
Let the type do the heavy lifting
Illustration budgets shrink and stock photography all looks the same. Typography is the one brand asset that appears on every screen — and most design systems still treat it as an afterthought.
2 min read
29 April 2026
Postgres is the default database, and it should be
Documents, search, queues, vectors for AI retrieval — one well-run Postgres now covers what used to take four products. The reasons to add a second database are fewer than they were.
3 min read
22 April 2026
Your AI assistant is part of your attack surface
Anything a model can read, an attacker can write to. Prompt injection is not a curiosity; it is the SQL injection of this decade, and it needs the same boring discipline.
3 min read
15 April 2026
Agents in the back office start with reconciliation
The most useful AI agents we have deployed this year do not talk to customers. They match invoices, chase exceptions and hand a person the three cases that actually need judgement.
2 min read
8 April 2026
The case against replacing your ERP
Most integration pain is process pain wearing a technical costume. How to tell the difference before committing to a migration.
2 min read
1 April 2026
Evaluations are the new unit tests
An AI feature without an evaluation suite is a feature you cannot change safely. Treat the eval set like test code, and most of the fear goes out of shipping.
3 min read
25 March 2026
TypeScript everywhere, and where it stops paying
Types across the whole stack have been one of the best trades in software this decade. But there is a point where the type system becomes the product, and the product suffers.
2 min read
18 March 2026
The quiet interface
Gradients, glass and bounce had their decade. The interfaces aging best in 2026 are the ones that get out of the way — and that restraint is harder to design than decoration.
2 min read
11 March 2026
What a technology roadmap owes the board
Three questions a roadmap has to answer before it is worth presenting — what stops, what starts, and what it costs to be wrong.
2 min read
4 March 2026
Retrieval before fine-tuning
When a model doesn't know your business, the instinct is to train it. Nine times in ten, the better answer is to hand it the right documents at the right moment.
3 min read
25 February 2026
Metrics before models
Why the first automation project should be the one nobody wants to demo, and how to sequence the rest behind it.
2 min read
18 February 2026
Design tokens are a contract, not a palette
Most token files we inherit are a list of colours with clever names. The useful ones are agreements between design and engineering about what may change, and what must not.
2 min read
11 February 2026
The boring stack is winning, and that is good news
A relational database, server-rendered pages and one well-understood cloud. The most reliable systems we saw last year were built from parts nobody would put on a conference slide.
2 min read
4 February 2026
Choosing an AI model is now a procurement decision
The leaderboard tells you which model is cleverest in general. It cannot tell you which one is cheapest to be right with, on your data, under your privacy obligations.
3 min read
Start with the two-week diagnostic.
A 60-minute review, then a fixed-fee fortnight. You leave with a costed roadmap and no obligation to continue.
Email
admin@yondertech.co.nz
Office
23A Dromorne Rd, Remuera
Auckland
Auckland