Every paradigm shift in technology has enabled people to do more with less.

Turing laid the foundation of modern computing by breaking the enigma — something that would have taken years of hand-calculation collapsed into an hour. IBM productionized that with mainframes that pushed time-to-solve down to minutes, though a single machine fills an apartment. Apple and Microsoft packaged all of it into a box on a desk, democratizing access to compute and forever changing the way we work.

That same shift is happening with coding agents. The cost of software creation has fallen to near zero — a single person can now build at the speed of an engineering team. Building is fun again because you no longer have to dread wrangling a pile of frameworks just to bootstrap something. And managers are becoming ICs again because they no longer lack the human bandwidth to build something big by their standards, and their management skills carry over — managing a team of AI agents draws on the same communication and planning skills as managing a team of humans.

I love the fact that we’re all reverting back to building stuff. Building is the single most important thing one can do for a business. Building is execution-based; it allows you to explore an option not known before. When the ground shifts underneath your feet and new ways to work with AI develop every week, there’s no clear playbook to follow, so a person with a bias to action is more likely to win out over a person who either takes too long to move towards a direction or apply an old playbook that might no longer be relevant or effective.

I find this bias of action super critical to hiring in a startup, and it’s also one of my personal gripes with overweighting a person’s years of experience or shiny credentials when making the first few hires. How people use their years of experience can go all the way from following a rigid playbook to using it to inform first principles — not everyone operates like the latter, which makes them a bad fit in a landscape where new methods are still being discovered and problems are still unsolved. One big example of this is to blindly hire middle managers based on whether or not they are from FAANG / big tech / consulting. Not all of these brand name companies have a work culture that’s necessary to survive the AI age; they often have middle managers who are not used to executing and adapting at a rapid pace, and this would become a huge culture shock when they transition to a startup where most of their time needs to be spent on the skills that they have not done much reps for.

To be clear, I don’t think managers will go away — middle managers who can’t multiply themselves to do more with less will. Since the cost of software creation is near zero, you in turn have more software creation — most of which is slop, but much of which is extremely valuable. Valuable software comes from understanding what it provides for the business and the end users. Because of that, you need people who fulfill two main purposes in the company:

1. Unify the company’s goal post + prioritize ruthlessly against a massive surface area created by the abundance of software products.

2. Think in terms of value provided by the software products and how to continuously provide more of them.

This forms the basis of the “modern manager” — not much different from the old way we do management on the surface, but it’s more about shipping products and less inventing unnecessary process and creating inefficiencies as a part of that.

The tricky thing is that this role is almost identical to what a CxO does — execute ruthlessly against a goal post to produce value. The modern manager has to be a replica of a CxO, which is an extremely high bar, and I don’t think most of the world has figured this out yet — we still have way too many middle managers. You only need them when the product surface area is so big that you’re competing against the giants of the world, e.g. FAANG: dozens of product verticals, each with dozens of features. Owning all of that as a single CxO would be chronic pain, so you hire a few extremely capable people to run each vertical on your behalf. That’s 1 layer below you — 1 manager per vertical, not 10. Yet most tech companies run one middle manager per pod, where each pod contributes to a vertical rather than owning it, and this is very wasteful.

Below that layer, I’d rather have super ICs. Middle managers aren’t needed when you have ICs who think beyond lines of code and own the business impact instead. Subject matter expertise still matters at some point, so specialists like PMs can help — but a super IC who can figure anything out doesn’t need someone in between relaying what they’re doing, because that IC is the one on the hook for getting it done. All it takes is checking in occasionally, being available for clarifications + questions, setting a deadline, and trusting they’ll get shit done no matter what. Every super IC you hire is 1 layer of org chart you don’t.