AI-Demented Development
I just watched Gojko and Dan’s unscripted contribution to goto;. Very entertaining – as always! I do love a bit of blunt Serbian humour to put things in perspective 😉
So, although I’m not a developer any more (and wasn’t a particularly good one when I was), I have been following the whole AI software development situation with great interest. Because if my job is helping the company get better value software out the door quicker, then I need to understand how best to do that. So even though I think I’ve played just about every role (except boss) in the software development lifecycle, I have not done so in the brave new world of AI.And I need to be able to say more than agent and MCP to be credible.
So, although I’m not a developer any more (and wasn’t a particularly good one when I was), I have been following the whole AI software development situation with great interest. Because if my job is helping the company get better value software out the door quicker, then I need to understand how best to do that. So even though I think I’ve played just about every role (except boss) in the software development lifecycle, I have not done so in the brave new world of AI. And I need to be able to say more than agent and MCP to be credible…
What I’m seeing at the moment, is that a handful of very gifted programmers – Gojko Adjiz, Kent Back, Martin Fowler, Dan North, Simon Wardley, Dave Farley, Peter Steinberger, etc – using their wisdom, experience to wrangle AI (to use Simon’s term) to increase their productivity dramatically. Virtually making them into one-man companies with the productivity of a whole team. However, this – and I could be wrong – invariably seems to be working on their own personal projects/products. And as Gojko said himself, you spent months wrangling AI and basically coercing it to write and think about code they way you do. And cause he’s brilliant, so are the results.
But I don’t work with people like them! 🤣 I work with the sweaty masses of good to very good developers, sitting in a development pipeline. Working on your own code, where you are the customer, requirements engineer, developer, quality assurance and often end user, creates instant feedback. There are no meetings to arrange, sign-off to get, process overheads. It’s all in your head. Which is probably quite exhausting! But also quite different from a large company like ours.
So my concern is – can this be made to work for the masses? Given that each context is different – Gojko laid great emphasis on changing things to suit your context, set-up and preferences. He also mentioned Wardley maps – and part of me thinks, yeah – this all moves from customized to productized. But I’m not sure.
Our company is currently talking about a x3 to x5 speed up. Am I completely naive? I can see how you, working on your own can achieve that – but not how we as a large company can. I personally think that 30% would be amazing. Am I completely naive?
Let’s just say that AI does deliver on productivity gains for large companies (not one-off quick wins and the like, but real gains). So by Q1 next year most companies have solved how to embed AI in the development and not be afraid of a CrowdStrike style meltdown – doesn’t that just create a level playing field and everything is the way it was before, with slightly more output? Or am I off the mark on this as well?