But “analysts”isn’t the end of abstraction, right? In the orgs you referenced (consulting firms, software cos, etc), analyst individual contributors are at the bottom of the career ladder.
Using the software ladder, you go up one rung and you have Tech Leads and Managers - the former is tasked solving high level technical questions while the latter deals with resource allocation, HR problems, cross functional collaboration etc. All of these are just different types of coordination problems. So wouldn’t the entry level just become these types of jobs instead?
For any software of sufficient complexity, i assume you can’t rely on a single agent (just like you couldn’t build Salesforce with 1 human engineer). You’d probably need a team of agents with specialized roles - maybe one that looks after architecture (Backend Dev agent), one that looks after the front end (UX Designer agent), one that talks to customers (PM agent). But the more profiles you introduce with diverging scopes, abilities, and goals (because LLMs are not deterministic, it’s not a given that agents will converge to the same conclusions!), the greater the coordination cost. So you need a human.
In other words, I’m not sure if I’m convinced that cheap intelligence automates away coordination cost. As soon as you introduce the concept of “agents working together”, you introduce the concept of “agents that are not fully aligned and require high level coordination”. It’s the same problem that arises in any human org of more than 2 people, but with agents instead.
So the job fundamentally shifts from execution to management or tech leadership - but the job still exists? And there still needs to be an entry level rung that trains people in how this works?
But “analysts”isn’t the end of abstraction, right? In the orgs you referenced (consulting firms, software cos, etc), analyst individual contributors are at the bottom of the career ladder.
Using the software ladder, you go up one rung and you have Tech Leads and Managers - the former is tasked solving high level technical questions while the latter deals with resource allocation, HR problems, cross functional collaboration etc. All of these are just different types of coordination problems. So wouldn’t the entry level just become these types of jobs instead?
For any software of sufficient complexity, i assume you can’t rely on a single agent (just like you couldn’t build Salesforce with 1 human engineer). You’d probably need a team of agents with specialized roles - maybe one that looks after architecture (Backend Dev agent), one that looks after the front end (UX Designer agent), one that talks to customers (PM agent). But the more profiles you introduce with diverging scopes, abilities, and goals (because LLMs are not deterministic, it’s not a given that agents will converge to the same conclusions!), the greater the coordination cost. So you need a human.
In other words, I’m not sure if I’m convinced that cheap intelligence automates away coordination cost. As soon as you introduce the concept of “agents working together”, you introduce the concept of “agents that are not fully aligned and require high level coordination”. It’s the same problem that arises in any human org of more than 2 people, but with agents instead.
So the job fundamentally shifts from execution to management or tech leadership - but the job still exists? And there still needs to be an entry level rung that trains people in how this works?
Awesome Alex. Grateful for sharing your thinking on a detailed level. Hope the new pay part of your enterprise motivates you to continue this path.
Good article ! How do you recommend longing the physical layer there are so many ETFs it’s hard to decipher