Blogs

The most consequential thing about browser-based design tools is not any feature. It is that the design, the comments on it, the prototype, the specification and the shared component library are all the same artefact at the same URL. Before that, a design was a file that someone The most consequential thing about browser-based design tools is not any feature. It is that the design, the comments on it, the prototype, the specification and the shared component library are all the same artefact…

There are two ways to put a generative AI feature into a product. You can assemble the platform yourself — gateway, orchestration, retrieval, guardrails, evaluation, observability — which is six months of work before the first feature ships. Or you can rent that platform and spen There are two ways to put a generative AI feature into a product. You can assemble the platform yourself — gateway, orchestration, retrieval, guardrails, evaluation, observability — which is six months of work before…

The appealing version of this idea is a pipeline where a requirement enters at one end and working software emerges at the other, with agents handling analysis, design, implementation, testing, review and deployment. The version that exists today is more interesting and considera The appealing version of this idea is a pipeline where a requirement enters at one end and working software emerges at the other, with agents handling analysis, design, implementation, testing, review and deployment.…

Retrieval-augmented generation is usually described in one sentence: look things up, then ask the model to answer using what you found. That sentence is accurate and it hides the entire problem, which is that "look things up" contains at least six decisions, each of which can qui Retrieval-augmented generation is usually described in one sentence: look things up, then ask the model to answer using what you found. That sentence is accurate and it hides the entire problem, which is that "look…

People describe using a capable language model as "talking to something that thinks", and then immediately worry that the phrase is misleading. Both reactions are reasonable. Something genuinely useful is happening between a question and an answer, and it is neither magic nor a l People describe using a capable language model as "talking to something that thinks", and then immediately worry that the phrase is misleading. Both reactions are reasonable. Something genuinely useful is happening…

A phrase entered the vocabulary and stuck: describing what you want in ordinary language and letting an AI produce the software. It has been called vibe coding, mostly affectionately, sometimes as an accusation. Either way, a genuinely new category of tool now exists, and a lot o A phrase entered the vocabulary and stuck: describing what you want in ordinary language and letting an AI produce the software. It has been called vibe coding, mostly affectionately, sometimes as an accusation.…

Almost everyone now uses a language model. Very few people have a working picture of what happens between typing a question and reading an answer — and without that picture, the technology alternates between seeming miraculous and seeming broken, with no way to predict which you Almost everyone now uses a language model. Very few people have a working picture of what happens between typing a question and reading an answer — and without that picture, the technology alternates between seeming…

Most people use ChatGPT as a very good chat box. That is a legitimate use and it is roughly a tenth of what the product has become. Underneath the conversation there is now a platform — custom assistants, connected data sources, file analysis, scheduled work, image generation and Most people use ChatGPT as a very good chat box. That is a legitimate use and it is roughly a tenth of what the product has become. Underneath the conversation there is now a platform — custom assistants, connected…

An AI coding agent is given a task in a repository and, some minutes later, presents a set of changes across several files with the test suite passing. It is a genuinely different thing from a code-completion tool, and the difference is not that the model is smarter — it is that An AI coding agent is given a task in a repository and, some minutes later, presents a set of changes across several files with the test suite passing. It is a genuinely different thing from a code-completion tool, and…