// Blog
In-depth technical articles from our team. No buzzword bingo, just substance.
For twenty years the IDE absorbed everything – panels, buttons, refactoring menus. AI coding agents are quietly shifting the centre of gravity back to the terminal, because they work in text, files and shell commands, not mouse clicks. Why the most composable interface is suddenly the most important one again – and where the IDE still wins.
AI systems are non-deterministic and they fail quietly: a wrong-but-plausible answer looks identical to a correct one in the logs. Classic observability measures error rates and latency, and misses the one signal that matters most: that the model has quietly got worse. Why AI systems need an observability discipline of their own.
Everyone frets about cloud and SaaS lock-in — almost no one budgets for the dependency on a single provider's model. Tie your prompts, tool-calling formats and fine-tunes tightly to one model and you turn a later switch into a rewrite, handing a vendor pricing power over your own margin. How model lock-in creeps in — and how to keep a cheap way out.
For years the standard frontend answer was: reach for a framework. AI code generation is rewriting the arithmetic, because a model scaffolds an entire UI in any stack in seconds — and that shrinks the ergonomic lead React and its peers won on. Is the era of the heavyweight framework ending, or is it stickier now than ever?
When the same model writes the code and the tests, the tests no longer verify anything — they encode the same misunderstanding a second time. A green bar then proves only that the code agrees with itself. Why testing in the age of AI has to be rethought: as the human-owned specification of correct behaviour.
When a model writes most of the code, the classic hiring signal – typing clean code fast at a whiteboard – measures the wrong thing. The scarce quality is no longer typing speed but judgement: reading, decomposing, verifying, and knowing WHAT to build. Why the interview is broken and how to fix it.
The reflex when building an AI agent is to make it more capable: more tools, more instructions, more context. Past a certain point, that makes the agent worse rather than better. The answer isn't a cleverer single agent but a system of specialists with narrow roles and clean hand-offs — along with the honest question of when the effort simply isn't worth it.
When a model writes most of the code, the bottleneck shifts from writing to reading. The decisive engineering skill of the coming years isn't producing code – it's reviewing someone else's code quickly and ruthlessly. Why code review becomes the core discipline, and how to build it.
Open-weight models have drastically narrowed the gap to the frontier. That reopens a question many considered settled: when is it worth running a model yourself instead of calling someone else's API? A sober decision framework beyond the ideology.
Three people with AI tooling now deliver what took an entire department five years ago. It's not just the tools that are shifting – it's the economics of team size. Why the next wave of productivity comes not from more headcount, but from less.
An AI feature that feels good in a demo tells you nothing about its quality in production. Without evals – systematic, repeatable measurement of model outputs – every team is flying blind. Why evaluation-driven development decides whether AI products succeed or fail.
Building an AI feature is cheap today. Running it for every user, every day, is not. Teams that don't understand their cost per request ship products with negative margins – and only notice when the bill arrives. A sober look at the economics behind every token.
The moment an application feeds a language model data from the outside world, it opens an attack surface classic security has no model for: instructions hidden in seemingly harmless content. Why prompt injection is the most stubborn unsolved security problem of the AI era – and how to contain the risk.
Many SaaS tools are nothing more than a pretty interface over an API you already pay for. Their whole business case rested on the cost of the glue code in between. AI is collapsing exactly that cost – and exposing which subscriptions are pure middlemen.
With version 7.0, WordPress integrates Gemini, Claude and ChatGPT directly into the system. It's meant to future-proof the CMS – and in the medium term it seals the end of plugins and themes. But it doesn't solve the real problem: the technological backwardness remains.
Time-to-Market was the central speed metric of the last decade. It's being replaced – by a KPI that sits further up the value stream and measures much more precisely whether a company can act.
For a decade the rule was: nobody builds back-office tools any more, that's what Enterprise SaaS is for. That dogma is tipping over – for three very tangible reasons.
Make or Buy was treated as a binary decision for decades. The most interesting software strategy of the moment is both at once: a bought platform with your own layer on top.
For years, data was a buzzword in strategy decks. With AI, it's become the hardest currency in competition – and most companies are giving it away without noticing.
For decades, custom development was a luxury and off-the-shelf software was a requirement. AI is reversing that – and suddenly software becomes a lever that actually sets companies apart.
Million-token context windows had everyone declaring RAG dead in 2025. They were wrong – but the classic RAG recipe is dead too. A look at the retrieval patterns that actually hold up in production today.
Prompt engineering was the buzzword of 2023. Three years on, the term barely describes what teams actually do. The real leverage sits in the entire context window – not in the prompt itself.
Per-user licences were the foundation of the entire SaaS industry. With AI, the model no longer makes sense – and the first vendors are starting to notice.
AI agents that read web pages, fill in forms and click their way through entire workflows have hit production quality in 2026 – at least for the right use cases. What works, what doesn't, and where the traps are hiding.
The EU AI Act has been live for high-risk systems since February 2026. Most companies underestimate what that concretely means – and overestimate what applies to them.
Everyone's talking about autonomous AI agents. But what happens when you actually deploy them into production systems? An honest look at what works – and what's still just marketing.
Legacy code does not just cost developer time. It slows product development, drives away talent, and turns every new feature into a risk. An honest accounting.
Building a pretty UI around ChatGPT and charging £29 a month – that no longer works. Why AI wrappers are not a business model, and what works instead.
Already 4% of all GitHub commits come from Claude Code – by end of 2026, it's projected to exceed 20%. What this means for developers, agencies and the future of software development.
Most teams build the app first, then the API. That's backwards. API-First Design flips the process – with massive benefits for parallelisation, quality and future-proofing.
Every development team knows technical debt. But very few have a strategy for reducing it systematically. We present an approach that works in practice.
Programming through natural language instead of syntax. Vibe Coding is more than a trend – it's a paradigm shift that blurs the line between developers and non-developers.
Perl, Delphi, ColdFusion – the graveyard of programming languages keeps growing. What determines whether a language survives? And which of today's popular languages will be irrelevant in ten years?
MCP is the new standard that allows AI models to access external tools and data sources. We show how to build MCP servers, when they make sense – and when you're better off without them.
We've used both AI coding tools in real projects – from refactoring to bug fixing to feature development. Here's what actually works and what doesn't.
43% of the web runs on WordPress – and that's exactly the problem. Why the CMS of the 2000s has no place in a world of headless, edge and AI-generated sites.
Seat-based pricing models, feature gating and bloated dashboards – the classic SaaS playbook no longer works. AI is replacing entire product categories faster than founders can pivot.