Surely satire right?
It makes sense to me… I use a very heavy framework that ensures my agent doesn’t lose context about the systems I develop. But that means every change goes through a big long pipeline and if all you want to do is change a few lines of code then maybe a junior dev is the right fit for that specific task?
Agentic coding would still have the context issues of changing code whether it’s AI or a human: somebody changed something, how do you record that for the next person. You either log it in memory or point it to the git PR, either way it needs to surface the changes.
So yeah, if AI is too expensive to code small problems for a given company than it’s too expensive for them period.
If you just want to change a few lines of code why not just, you know, change them?..
The hard part isn’t changing the lines of code it’s finding them. An agent can do it in the background for you while you work on other things rather than spend 10-15 minutes hunting for the exact line you want
For those asking / pronouncing this has to be satire, perhaps. But not for long. AI is still not making a profit. So whatever it costs today at the growth-at-all-costs subsidized rate, think how much more expensive it will be when investors start insisting on profit after market consolidation*.
Because if you think there is a competitive barrier to entry for smartphones, operating systems, CPUs, and streaming services, you ain’t seen NOTHING yet
They’re spending $3 per $1 of revenue. The price per token will rise dramatically.
I suspect it’s even worse than that.
A Claude Max subscription is $200 a month, which is roughly $7 a day. I’m forced to use Claude Code at work, and I frequently run the /usage command out of morbid curiosity to see how many tokens in wasting. I’m not exactly a power user, but even with my bare-minimum usage I typically burn about $50 of tokens per day — so roughly 7 times that $7 a day figure.
And that’s just based on Anthropic’s official per-token pricing, which itself is almost certainly subsidized… so it’s likely I’m costing them something closer to $20-$30 for every $1 of revenue. And again, I’m only using it for the bare minimum. I know people with much higher usage than me.
Similar boat, lots of people on the loss-leading phase, ‘the first hit is free/cheap’ is in play and I know development organizations that are explicitly designing hard dependencies on hosted AI as the ‘foundation’ of their processes. They are going for mainframe-style lockin where customers are too scared to change when it gets too expensive.
I know one organization that at least is being more careful, their LLM usage is only based on whatever they can run indefinitely on-premise without sweating future traps around pricing changes.
Junior Developer doesn’t boil oceans to do their job
Not with that attitude.
We don’t usually take into account what employees do to nature on their spare time.
In a way, we would have to equate all the money spent on employees and what it is used for.
Getting rid of an employee would surely be a net gain for the environment, right? If an LLM could offset 50 employees, perhaps that would be neutral in that sense.
This doesn’t make a lot of sense, esoecially morally. We can’t really blame the company for what employees do out of work, can we?
