Hi there 👋

The lonely architect is on a quest to make apps1 with ease. In today’s software world there are numerous options when it comes to building apps and finding the best approach has proven to be a tiresome endeavour with no end in sight. At the same time, the world seems to move forward with no regard to the lonely architect’s plight; churning out new solution after solution. However, I ask of the world to show some patience, have a good look at the existing solutions and think of where the gaps actually are. ...

Before the recent AI buzz, we had the ML (Machine Learning) buzz. I believe thinking in terms of the latter is a better way to assess the long term usefulness of the technology. My friend Thanh, who is working on building solutions that involve AI powered code generation shared with me how he thought of the system he is building: i think of the AI as an intern … eventually the system should know enough to work independently ...

One could think of coding as a simple process of the programmer producing the code. But, what if we want to focus on the fact that the code and the ideal specification are never identical. In this case, we could think of the programmer as introducing noise. In other words, when software is coded by hand, the ideal specification becomes muddied by the programmer in some ways before it becomes code. However, as the programmer gets better (in ability and in his own understanding of the problem), the noise he introduces reduces, eventually allowing the code to perfectly capture the specification. ...

With the advent of AI coding tools1, a war is now on. It is not about productivity or efficiency - it is about the soul of programming itself. Let’s begin with the following argument that is often made: Why know the details ? Isn’t the high level requirements all that matters ? I would like to answer it indirectly. By first reframing it in the following terms: doesn’t the black box behaviour of a program unambiguously define it ? ...

Current AI systems are very primitive. For them to be truly useful long term, they will need enhancements. I want to talk about one which is hardly available in the current widely deployed systems. I will be discussing specifically in the context of coding agents. The points are likely very applicable in most other areas too. That feature is often named as continual learning in the literature. If we are to compare a human with an AI system today, this is likely the biggest functional difference. ...

How LLMs Learn — And What That Means To set the stage, let’s think about how LLMs (Large Language Models) - the predominant AI system in use to do software development today - work. A model is a mathematical function defined through billions of parameters. These parameters are tuned based on training data. The goal of training to make the function capable of doing a good job of predicting the next token, given a set of input tokens. ...

As AI starts to leave the lab and diffuse into the workplace, AI slop is increasingly becoming a big problem that organizations have to deal with. Let’s define it first. AI slop is any work product that contains no new intelligent information. It is just a retelling of what is already known. It is often the result of unintelligent prompting and/or lack of context for the AI to draw on. ...

There is no need to fear AI. No, this is not an article motivating you to work hard to keep up. Sounds interesting? Read on. It is not a stretch to say that there is a whole industry out there in cyberspace, dedicated to reassuring that AI is a bubble. This industry exists because there are a lot of ears waiting to hear just that. But what if there is no need to feel threatened in the first place? That, irrespective of the technical details of where the limits of AI capabilities lie, at a fundamental level, we are not in a competitive relationship with AI. ...

Talent pipeline is important for an organisation. Imagine the software industry. Junior engineers learn their craft through the mentorship they receive from seniors. Today’s juniors are tomorrow’s seniors. This is what we call the talent pipeline. It is one of those things where everyone has to do their small part. If they don’t, it may not seem to matter in the short term, but as a whole it can be an industry killer. ...

The Emergence of AI-First Programming A new kind of programming is emerging — one that is AI-first. Humans will supervise AI work and jump into details where needed. The AI handles the rest. New people entering the field will also enter it AI-first. They’ll build and publish entire apps in hours. Access to AI tools would be a standard part of universities’ infrastructure, just as a library of books is today. ...