AI, Semantics and the Value Threshold
There’s been a lot written about AI in the last couple of years, and I slightly hesitate to add to that because I’m not sure I can add much to that discussion. But this caught my interest. There’s an article on Simon Willison’s blog (which is engagingly still called a Weblog) about how he uses LLMs for coding.
The opening paragraph resonates. We’re seeing a lot of hype, but not experiencing the life-changing results that others seem to be describing. We still have concerns. So finding someone who doesn’t rhapsodise about LLMs is good. He goes on to make a lot of good points about things like setting reasonable expectations and managing context. But where I got interested is when he talks about ‘weekend projects’.
A weekend project is a little ‘wouldn’t it be great if…’ project that takes no more than a couple of days to throw together, and where quality isn’t necessarily an issue. And in a lot of ways, it’s those qualities that make these projects great testing grounds for what AI can do. When the project is small, providing all the context becomes feasible. When the outcomes are not critical, quality control is not such a concern. Which means you can leverage the efficiencies offered by pair programming with LLMs without some of the risk, and substantially cut down your development time.
Which – and this is the cool bit – moves the value threshold*. There’s a line, on one side of which are the projects that clearly have enough value to be worth investing time in. On the other side of the line (the sad side), are the projects that are never going to make the cut. Some great ideas, but when it comes to it, not valuable enough to warrant the effort involved.
Like most dev teams, we have more ideas than we have time, so there are quite a few little gems languishing in the idea graveyard on the sad side of the value threshold. And if our LLM friends can help resurrect a few of those by cutting the effort and therefore moving the threshold, wouldn’t that be nice?
*Footnote
Semantics being a rabbit hole I’ve been known to fall down…. we talk about a “value threshold” because there doesn’t seem to be a good term for that cut-off point between things that are worth doing and things that – be honest – you’re never actually going to do. There’s lots of related options like ‘point of diminishing returns’ and ‘cost/benefit analysis’ but none are quite the right fit. Incidentally the origins of the word “threshold” date back to when straw (thresh) was considered an upmarket floor covering, which was all good except that it had to be held down in the doorways to stop it blowing around. Haven’t we come a long way?