Skip to main content

From the Frontier

Insights on AI-native engineering, frontier AI workflows, and building production systems at speed.

Painted illustration of an engineer with a tablet walking past rows of server cabinets in a sunlit glass-walled hall, hills beyond the windows

production-systems · security

PHP2Shell: What CVE-2026-109618 and CVE-2026-109619 Mean for Your PHP Stack

Two critical flaws in the PHP runtime allow unauthenticated remote code execution, even on a Hello World app. What is affected, what attackers are doing, and what to patch today.

Read post
Painted illustration of two people in navy jackets leaning over a large pale chart on a sunlit table, one pointing out a route, with moored boats beyond the windows

consulting · enterprise

How to Choose an AI Engineering Partner (Without Getting Burned)

Most AI consultancies sell demos. Here's a practical framework for evaluating partners based on what actually predicts project success — and the red flags that predict failure.

Read post
Painted illustration of a grower in a navy jacket at a table with a record book of rising colour bars, facing fruit-laden orchard rows and a white marker pole

enterprise · methodology

Measuring ROI on AI: The Metrics That Actually Matter

Most AI investments can't prove their value because nobody defined success upfront. Here's how to build a business case, pick the right metrics, and know whether your AI project is actually working.

Read post
Painted illustration of a woodworker in a navy apron at a sunlit workbench beside planks and a yellow board, a large idle machine standing behind

methodology · consulting

When Not to Use AI: A Checklist from People Who Ship It

We build AI systems for a living, and we regularly talk clients out of using AI. Here are the situations where traditional software, human processes, or doing nothing will outperform any model.

Read post
Painted illustration of an inspector in a navy jacket, seen from above and behind, carrying a tablet across a sunlit courtyard toward a row of red and grey doors

production-systems · security

Vibe Coding Security: What AI Coding Tools Don't Check

AI coding tools like Cursor, Copilot, v0, and Bolt let you ship fast. But the code they generate has predictable security gaps. Here's what to look for — and how to fix it before someone else finds it.

Read post
Painted illustration of two boatbuilders in navy jackets conferring over a tablet beside a finished dinghy on trestles, with masts and water beyond the open workshop doors

production-systems · consulting

Startup Speed, Enterprise Quality: You Can Have Both

The idea that you have to choose between moving fast and building properly is outdated. AI-native engineering delivers both — and here's how.

Read post
Painted illustration of a cook's hands slicing on a sunlit kitchen counter, small glass bowls of prepared ingredients set out within easy reach

ai-engineering · methodology

The Developer Experience Gap: Why Your Engineers Hate Your AI Tools

Most enterprise AI tooling has terrible developer experience. If your engineers aren't using the AI tools you bought them, the tools are the problem — not your engineers.

Read post
Painted illustration of two potters at a sunlit studio bench covered in small coloured glaze test pieces, one holding a fresh piece up to the light near the kiln

ai-engineering · claude-code · codex · methodology

What AI-Native Engineering Actually Means (And Why It Matters)

AI-native isn't about bolting ChatGPT onto your existing workflow. It's a fundamental rethink of how software gets built. Here's what we've learned shipping production systems with Claude Code and Codex.

Read post
Painted illustration of a homeowner's hands at a sunlit kitchen table, one holding up a single key over unrolled house plans beside a yellow mug and an open window

domain-intelligence · enterprise

The No-Code AI Trap: Why Low-Code Platforms Create High-Cost Problems

No-code AI platforms promise democratisation. They deliver vendor lock-in, limited customisation, and costs that scale faster than your business.

Read post
Painted illustration of an engineer in a yellow hard hat on a rooftop holding a large plan over a city taking shape below, cranes at work and a river beyond

ai-engineering · methodology

The Software Engineer of 2027 Doesn't Look Like You Think

Software engineering is changing faster than any time since the internet. Here's what the role looks like in two years — and how to prepare.

Read post
Painted illustration of a person in a navy jacket reading a plan beside the back of a large translucent clock face, new brass parts laid out on a bench

enterprise · case-study · ai-engineering

Your FileMaker System Isn't the Problem — Your Migration Plan Is

FileMaker served your business for years. Now it's holding you back. Most migration plans fail because they treat it as a rewrite. Here's why extraction-first wins — and what an honest audit actually looks like.

Read post
Painted illustration of a woman in a navy jacket seated at a long sunlit desk in an arched archive hall, a small stack of folders beside her and shelves behind

ai-engineering · methodology

Forget Prompt Engineering — Context Engineering Is What Matters Now

Prompt engineering got all the attention, but the teams shipping real AI systems have moved on. Context engineering — controlling what the model sees, not just what you ask — is the discipline that separates demos from production.

Read post
Painted illustration of a dispatcher in a navy jacket walking beside a roller conveyor carrying bright parcels toward a sunlit open dock door in a dispatch hall

testing · production-systems

Shipping Fast Without Breaking Things: Our Testing Philosophy

Speed without quality is just recklessness. Here's how we ship production systems in weeks while maintaining the kind of test coverage that lets us sleep at night.

Read post
Painted illustration of a signaller in a navy jacket at a desk of small coloured levers, looking out through tall windows over a sunlit rail yard

ai-engineering · enterprise

AI Agents in Business: Hype, Reality, and What Actually Works

AI agents are the hottest topic in tech. Most of what you've heard is hype. Here's what actually works in production today — and what's still science fiction.

Read post
Painted illustration of a weaver seen from above at a sunlit table, charting an old red rug's pattern in colour blocks beside spools of yarn and a small loom

enterprise · case-study

Legacy System Modernisation: The AI Shortcut Nobody Talks About

AI can read, understand, and translate legacy code faster than any human team. Here's how we use it to modernise systems in weeks instead of years.

Read post
Painted illustration of two florists in navy coats at a sunlit bench, one cutting ribbon with scissors while the other holds a bunch of yellow and red flowers

ai-engineering · methodology

Your Team Doesn't Need AI Training — They Need AI Tooling

Stop sending your team to AI workshops. Instead, embed AI directly into their existing workflows. The best AI adoption happens when people don't even notice it.

Read post
Painted illustration of a harbourmaster in a navy and yellow jacket at a glass wall, looking out over a calm harbour sheltered by two stone breakwaters and a lighthouse

production-systems · enterprise

Security for AI Systems: Beyond the OWASP Top 10

Traditional web security doesn't cover the attack surface of AI systems. Prompt injection, data poisoning, and model extraction need new security patterns.

Read post
Painted illustration of a worker in a navy jacket walking a sunlit mill walkway past ranks of orange bins, a barred gate and round lamps ahead

production-systems · ai-engineering

Data Pipelines Are the Unsexy Foundation of Every AI System

Everyone wants to talk about models and prompts. Nobody wants to talk about data pipelines. But pipelines determine whether your AI system works in production or just in demos.

Read post
Painted illustration of a worker in a yellow cap crouching to set the next slab of a garden path leading to an open door, blocks stacked nearby

enterprise · consulting

Five Enterprise AI Mistakes We See Every Month

After working with dozens of enterprises, the same mistakes keep showing up. Here are the five most common — and how to avoid them.

Read post
Painted illustration of two people midway across a slender footbridge at golden hour, one pointing across the water, the other resting a hand on the orange handrail

consulting · methodology

Why We Don't Bill by the Hour

Hourly billing incentivises slowness. We charge for outcomes because AI makes the 'hours spent' metric meaningless — and our clients prefer it that way.

Read post
Painted illustration of a printer's hands in navy sleeves with yellow cuffs holding loupes over two colour-swatch proof sheets on a sunlit table scattered with pins

domain-intelligence · testing

You Can't Improve What You Can't Measure: AI Evaluation Done Right

Most AI systems ship without meaningful evaluation. Here's how to build evaluation frameworks that actually tell you whether your AI is working — before your users do.

Read post
Painted illustration of an engineer's hands with a straight-edge and yellow pencil over a plan of matching coloured modules on a sunlit drafting table

production-systems · methodology

AI Doesn't Create Technical Debt — Bad Engineering Does

The fear that AI-generated code creates technical debt is backwards. AI-assisted code with proper engineering practices produces less debt than traditional development.

Read post
Painted illustration of a craftsperson's hand setting a chisel onto a wooden board among shavings and hand tools on a sunlit workshop bench

ai-engineering · case-study

The Internal Tool Renaissance: Why AI Makes Build-Your-Own Viable Again

For a decade, buying SaaS was cheaper than building. AI has flipped that equation. Internal tools are back — and they're better than anything you can buy.

Read post
Painted illustration of a lock-keeper in a navy jacket walking along sunlit cream stone copings beside dark canal water, a stepped flight of orange lock structures rising beyond

enterprise · methodology

How Long Does an AI Project Actually Take?

Everyone asks how long an AI project takes. The honest answer is 'it depends' — but here's a practical framework for estimating timelines that actually holds up.

Read post
Painted illustration of a white-bearded horticulturist pointing along a sunlit greenhouse bench while a younger colleague in a rust coat listens, holding a colour chart

domain-intelligence · methodology

Why Domain Experts Should Control AI Systems (Not Engineers)

The biggest mistake in enterprise AI is treating it as a pure technology problem. The best AI systems are built by domain experts with engineering support — not the other way around.

Read post
Painted illustration of a bookbinder in a yellow apron standing at a long sunlit bench, handling a fresh book beside neat stacks of finished blue and yellow volumes

claude-code · codex · case-study

Claude Code and Codex in Production: What We've Learned After 50 Deployments

After shipping 50+ production systems with Claude Code and Codex, here are the patterns that work, the pitfalls that don't, and why the tool matters less than the methodology.

Read post
Painted illustration of an engineer in a navy coat comparing three small contrasting block models on a sunlit wooden workbench, notes in hand, long shadows on the floor

ai-engineering · methodology

Three Prototypes in a Day: How AI Changes Architecture Decisions

When prototyping costs hours instead of weeks, you can explore multiple architectures before committing. This changes how teams should make technical decisions.

Read post
Painted illustration of a tailor in a navy waistcoat and yellow tie at a sunlit cutting table, with trays of buttons, shears and finished jackets hanging behind

enterprise · methodology

Build vs. Buy: When Custom AI Beats Off-the-Shelf

The build-vs-buy decision changes fundamentally when AI collapses development timelines. Here's our framework for when custom beats vendor — and when it doesn't.

Read post

Want to work like this?

We write about what we ship. If you like how we think, imagine what we can build together.