"I’ve now seen dozens of companies struggling to put AI to work writing code, and each one has moved through five clear tiers of automation. That felt familiar."
Software development is transforming incredibly quickly. This guide, although it looks a bit silly at first glance, is very aligned with what I’ve heard from friends and seen first-hand.
“I’ve now seen dozens of companies struggling to put AI to work writing code, and each one has moved through five clear tiers of automation. That felt familiar, and I realized that the federal government had been there first – but for cars.
In 2013, the NHTSA created…
"I’ve now seen dozens of companies struggling to put AI to work writing code, and each one has moved through five clear tiers of automation. That felt familiar."
Software development is transforming incredibly quickly. This guide, although it looks a bit silly at first glance, is very aligned with what I’ve heard from friends and seen first-hand.
“I’ve now seen dozens of companies struggling to put AI to work writing code, and each one has moved through five clear tiers of automation. That felt familiar, and I realized that the federal government had been there first – but for cars.
In 2013, the NHTSA created the five levels of driving automation. This was helpful, because while the highest level at the time was only level 22, it let everyone have a common language for both where things were, and where things were going.”
What follows is instructive; some of the interim steps are waypoints, but 4 and 5 seems to be where a lot of mainstream software development is going. We can thank Claude Code for this; many of these changes took place last year, with some pre-work laid down by the AI vendors and various startups beforehand. My guess is that everything will have changed again by the end of the year.
In the hands of senior engineers who are getting their heads around how these tools work, AI coding is starting to work as advertised. There’s a reason why engineers who have been coding for decades rave about it. But we’ll also see some really bad code (particularly in enterprise organizations) and some high-profile failures. There’s something I’ve called The Mythical Claude Code Agent Month (I need to find a pithier name), where in response to process and culture failures that are hampering its software development, an organization just decides that it needs to add more AI. And hallucinations, bias, and model poisoning are all real things.
In response to this, and because a lot of engineers are ideologically or otherwise opposed to AI, I think we’ll begin to see explicitly artisanal software companies emerge. In some industries, particularly highly-regulated ones, we’ll also see new kinds of trust certificates emerge to prove that (regardless of how it was built) software performs well and doesn’t leak private information.
Anyway, those are all knock-on effects. As of now, this is where mainstream software development is likely going. This isn’t an endorsement, necessarily — but it is a good-faith observation.
[Link]