Humanity is self-deprecating
A year ago agents were good, but comparing them today is a folly. Today, agents effectively replace junior developers and companies have realized that.
I am not arguing for or against this; it is merely a fact. Hiring of juniors and even mid-level developers has slowed greatly, and stopped in some places.
Who's going to be the seniors of the future if we're not training any junior-to-mid developers? Well, companies don't care about that. They care about the money, and less people means less money spent.
That's about as deep as it goes - they don't really care about what it means a year or a few years from now. They care that it makes this quarter's earnings better, and that the CEO can get the bonus they want.
Then, what about the seniors? The ones that know-the-things, that have maybe been there for a while, and can reason about things in the way that agents can't?
Their days are numbered, too, and I suspect it's closer than we expect. Agents are improving at a rate faster than we can anticipate, and this rate is only growing. The capabilities model-to-model is vastly different to what it was 2 years ago. I'm not only talking about benchmarks. The off-bench performance is very different between model generations too.
So when agents can manage millions of tokens of context without a hitch, use it effectively, and achieve near-perfect ability to navigate, understand, and reason about even large codebases, where do the seniors fit in?
Sure, seniors can reason. Though if an agent can do everything it does now, but way, way better, that's just a junior agent that's now a senior, and now replacing seniors.
But, the managers! The managers, the seniors that manage people! They're safe, right?
No. Agents will manage agents. At this point, the landscape will have shifted considerably and be unrecognisable from today. We won't be prompting them, we won't be building skills, and we won't be optimising random things to squeeze better performance out, because the agents will do all of that, and better.
This is only the start, too. Replacing seniors is early-days for this tech. 20 years in the future we'll be in a different world.
I am honestly not sure of what to do about that. It's a sort of paralysis, perhaps. Right now, things are "okay," and "good," even, for seniors. But in a few years this will change. Then what?
No idea. Get good at managing agents now. That seems to be the only thing that's at least somewhat sensible. Though when agents eat that too, I don't know where to go.
But, if all we do is manage agents, the skills we once had to do the work ourselves atrophies and we only increase dependence on the agents themselves.
What would happen if all worthwhile agent providers bumped the price of their agent by 1000x tomorrow? Sure, we'd complain, and then what? We'd suck it up. The lucky ones that can afford it would, at least, and the rest would be paralysed.
Somehow, I think we're already past the point of no return, where the loss of this tech for a massive number of developers effectively means that work stops for them. We're directly dependent on the tool that's replacing us.
We're writing blog posts, developing wonderful new tech to make agents more effective, work better for us, and let us push code and solutions faster. Well, we're only helping the tech replace us faster. Ironically, the faster it replaces us, the faster we help it replace us.
Somehow it feels like we're on the cusp of an economic and employment apocalypse.
Humanity is self-deprecating.