My Mother was a Computer

Computers used to be women. That is “computer” used to be a job title for people, almost exclusively women, who performed complex, lengthy calculations required for applications such as the physics of space travel. These women used hand-cranked mechanical calculators, slide rules, and tables of logarithms.

Almost overnight their jobs were transformed by the advent of the mainframe computer. They were asked to put away their slide rules and, instead, learn to program those mainframes using FORTRAN.
 
If we could be transported back to the late 1950s, we would be stunned by the preternatural skills of those women. The accuracy and adroitness with those unwieldy and unforgiving tools. Think of the stress of doing computations that, if inaccurate, could lead to a rocket blowing up on the launch pad. Think of all the processes, the doing, the checking, the reviewing. The fact that those skills and processes became obsolete overnight make them no less impressive.
 
This, I think, is the closest analogue to what’s happening in software development today.
 
There is much heat generated by comparisons between compilers and generative AI. It is fair to point out that compilers are deterministic while LLMs are stochastic. I don’t think this vitiates the comparison, though. I wrote lots of Z80, 68000, and 8086 assembly language code in the past. But I’d struggle to write anything meaningful in those languages now. A compiler does that job better than me. Similarly, I’ve not issued a syntactically correct git command in about seven months. I just ask an agentic harness to “get the latest code”. 8086 assembly language and git commands have gone the way of the slide rule and tables of logarithms.
 
The mathematician Terence Tao posits a similarity between the industrialisation of food production and AI. The former resulted in cheap food for millions of people. While nobody would argue that there’s nobility in food poverty, it is true that we now have concerns around obesity, diabetes, and exercise that would have been difficult for a farmer in the nineteenth century to comprehend. Imagine trying to explain to someone who spent sixteen hours a day in backbreaking labour that you need to go to the gym to lift some weights. Tao thinks we could end up in the same place with cognition. We may need to think about how we use our minds in a world where we’ve delegated deep thought to machines. Calorie counting and gym sessions for your brain.
 
To me, this is the crux. The corollary of using AI is not that you necessarily become intellectually passive. AI can be used to have the exact opposite effect. AI can be used to challenge your thinking. On a basic level, asking an agentic harness to generate questions for you to answer to flush out fuzzy thinking or contradictions is such an example.
 
I’m not confident of much when it comes to the rapidly evolving world of agentic AI. One thing I am confident about, however, is that using AI to avoid deep thinking is a bad idea. Using AI as an insanely knowledgeable, if naïve and sycophantic, collaborator that forces you to think hard about what you’re trying to achieve is a good idea. Using AI to tailor educational material to exactly match your angle, learning preferences, current level of comprehension, and intended target is better still.