AI copilots for the eternal tinkerer

If you write code, you're a programmer. If you start ventures, you're an entrepreneur. Simple, right? But what are you when you're obsessed with experimenting with every new framework that hits Hacker News?

I've spent over a decade being that person – the eternal tinkerer. Back in university, I thought chasing every trending technology would give me an edge. Now, with the rise of AI and LLMs, that same curiosity has evolved into something far more powerful.

Here's the thing: While twitter is flooded with predictions about AI replacing developers, I'm seeing something different. The new tools are insanely awesome for starting new projects. After building documentation platforms, Xero tools and new backend frameworks, I've discovered that AI isn't our replacement.

Coding with AI by your side is like working with a really terrible Junior engineer. One that doesn’t learn, writes a lot of code, often goes down a rabbit hole. Creating prototypes has never been more fun or easy as a result.

Towards higher and higher levels of code

The last 40 years of progress in computing has been all about a shift from low level abstractions to high level conceptual solutions.

Solutions built for specific computer architectures were replaced with general purpose languages and better compilers that could target multiple architectures.

Languages became business oriented. COBOL came with ‘human’ friendly syntax closer to stakeholder language. Computers got fast enough for dynamic languages.

Platforms for massive distribution became a thing, java, the web, mobile. We saw domination by organisations that could monopolize a platform. Big tech companies go through multi billion dollar shifts to go after speculated new platforms like virtual reality.

But in October 2021 the next big thing actually happened, and that was the release of github copilot. After a couple years with this AI sidekick, we can clearly see this as yet another shift on the continuum of focusing on the customer pain first. Prototyping has become easier than ever before. Even people who have never written a line of code can build adequate systems for testing a solution.

AI is great for exploration

Some thought leaders might tell you that AI is replacing us, that you should buy their course and change careers. The reality is that new AI tools can be great for exploring a problem space and building prototypes. When software actually becomes more valuable and legacy, then the effectiveness and usefulness of AI falls off a cliff. We still need humans in the loop to guide agents towards sustainable extension of legacy codebases.

Software runs through this life-cycle forever
All software runs through the same cycle of exploration and sunset into legacy. Agents will not change this reality. Starting something new, the most important thing to do is get your solution in front of the people who are trying to do the job with it. If you get it right, the application moves on, code runs for months and years. As the codebase grows, it shifts towards a state of optimization and happy legacy. Once that legacy becomes too brittle, and everyone who originally wrote it moves on – you begin the rebuild. You pull out threads from the wonderful legacy system and build little components to explore solving the problem again.

Frameworks have a similar value lifecycle. In the beginning they provide guard rails to your development. They prevent solving the boilerplate problems like authentication. Within a year, most of that value is gone. You start to build your own abstractions. After 5 years, that framework is probably completely unrecognisable from an application written with a custom framework.

We should care about legacy software. The expected future value of a technology is proportional to the length of time it has existed so far. This is also known as the Lindy Effect. Agents so far have proven themselves not very capable enhancing legacy software.

Where does current AI fit into the paradigm?

Keep your agents focused on rapid prototyping for now. Use them to work very closely with the designers, customers and experts wherever you are applying your craft. We can learn faster than ever about the ‘job to be done’ by leveraging our new prototyping super powers.

As of September 2025 – I would say always be experimenting. You should dispatch an agent on a legacy code base, you should craft your prompts and context to bring in the right type of changes to your legacy code.

The future is exciting

This old curmudgeon of a developer is enthusiastically trying all of the AI tools. I still keep a keen eye on hacker news and twitter for the latest and greatest. I am approaching the new tools and techniques with excitement and skepticism.

If you are just entering the industry, or thinking about it. Do take up the AI tools, and experiment with the new shiny thing. But also, do not take serious account of AI grifters and doomers who might say that the end of our industry is just around the corner. We can have fulfilling and fun careers building really valuable software with copilots and agents at our side.

Technological progress has not been a constant in the world, but throughout my life and yours it will be. We either ride the wave, or we take up new careers. Join me on the adventure, I really look forward to what the future holds for our industry.

As always, if you’re interested in semi-non-regular emails about my thinking, you can sign up to the RSS feed on this site or put your email in the box below for updates.

Peace & happy coding.