A friend and I recently discussed how the qualitative experience of coding has fundamentally shifted in a key way these past few months. Engineers morphed from early-2025 snickering at MBAs vibe-coding themselves simple demos that’d surely never reach production to late-2025 embracing of AI as an integral part of professional software engineering.
We have passed the event horizon. Pandora’s box has been opened. The genie is out. There is clearly no going back.
What may not be obvious to everyone is that this shift also killed the future of the individual contributor. I’m not here referring to whether AI will replace ICs. I’m instead saying the fundamental software job of “individual contributor” is permanently going away due to AI, whether or not (actually, especially if) you’re better at coding than AI.
Amongst other things, managers and leaders in software teams:
Set priorities
Decide amongst disparate architectures
Match tasks with employees’ talents and interests
Resolve conflicts
Give feedback
Maximize output of the given team they manage
The work of managers is meta-work: it’s not building the thing, it’s pondering and tweaking the machine that builds the thing. A great day of management isn’t “I laid 350 bricks today”; it’s “I helped unblock Bobby, who laid 350 bricks today.”
As AI gets increasingly capable, and its adoption becomes less a matter of debate and more a matter of pacing and strategy, the halcyon days of the IC are over. Not because AI codes better than you, but because maximizing your productivity necessitates focusing your time on all the things that are, at the end of the day, manager tasks.
Consider the things I spent the past month doing while coding on Superphonic:
Priorities. You’d think having to 10-12 AIs working in parallel would mean you could think less hard about priorities (“You could get so much more done! Why not do it all?”), but I find the opposite. Back when I coded all of 80 WPM, I moved slowly enough to reflect on priorities when getting a coffee or waiting for the build to finish. Now the assembly line of PRs whizzes past so quickly I need to constantly have ideas on what’s best to do next.
Architectures. Because the labor is (nearly) free, I can now spawn two different architectural approaches at once and decide amongst them either by running them or by inspecting the resultant code. With human employees, you’d need to be pretty careful before spinning off two teammates to essentially compete on different architectural approaches. These interpersonal dynamics don’t exist with AIs, which opens the door to being far more empirical when choosing between architectures. If you’re doing it right, this means you’ll be making architectural choices far more frequently than you used to.
Strengths matching. If you, like me, pay for multiple AIs, you’ll have this in spades. You quickly get a sense of which AIs would be good at what tasks and start allocating tasks accordingly.
Resolve conflicts. There’s a lot of “he said / she said” when you set two AIs onto the same thread and ask them to critique each other’s input. This becomes especially interesting when you don’t know as much about the domain as your AIs. It’s exactly what every executive is faced with all the time: discerning between the vociferously-defended, trenchantly-opposed viewpoints of experts who know far more than the executive.
Give feedback. I know on one hand that AIs don’t learn from feedback, but I still can’t resist the desire to “teach them how to fish.” More fruitfully, I update each AI’s custom instructions when I want them to behave differently. More often than not, as with human employees, this advice is only sometimes followed.
Maximize output. When I manage teams, I often stress about how to keep the task queue full and the employees unblocked. This now happens all the time with me and AI. Much of my day is now spent frantically figuring out parallelizable tasks for multiple AIs to tackle simultaneously. This problem only exacerbates as each AI gets better.
Whether you wanted it or not, your job is now managing a fleet of AIs. You’ll spend your days doing exactly the above, not thinking about implementations, finding bugs, or debating architectures with coworkers.
This doesn’t have to be your job, of course. If you’re independently wealthy and don’t need gainful employment, you can code all day, basking in the eternal sunshine of a mind spotless from the advent of AI.
But overwhelmingly, for the world of the employed, forces of market competition will compel you to manage fleets of AIs if you want a job at all.
Perhaps you’ve always wanted to be a manager. The above is then great news: I’ve basically ensured your management future regardless of your talent for it, with a team positively obsequious to your every brilliant utterance.
In fact, one could say managing a fleet of AIs is better than managing humans in many ways. To wit:
You never have to write performance reviews or argue with peer managers about various employees during hours-long “calibration meetings.”
You don’t need to deal with petty jealousies and Lincoln-esque Team of Rivals situations.
You never spend your entire day in back-to-back, double-booked meetings. (My longest ever meeting, funny enough, was an 8-hour video call for, of all things, team performance calibrations at Meta).
You never get subpoenaed for court appearances regarding allegations of bias or misconduct, whether yours or others’, baselessly for financial gain.
You don’t need to hound your AIs to take mandatory SOC2-compliance training or remind them incessantly about what it takes to maintain HIPAA, get reimbursed for travel, establish attorney-client privilege in emails, or obtain a temporary cardkey for entry during holiday hours.
It might currently feel like managing a team of barely-competent interns, but it’ll soon feel like managing a team of very high performers, each better, faster, and smarter than you. Yet beholden to your every whim.
What’s wrong with this, you say? Is it not utopic?
The world we’re forced headlong into may well be your idyll, in which case you should skip this section and just pass Molochinations along to a friend for mutual a-/be-musement.
Half-kidding self promotion aside, I’d here proffer a brief eulogy for what we’ve given up in this transition to all becoming managers whether we wanted to or not.
Meta-work, the type managers do all day, can be very fulfilling if you’re quite zen about the indirect long-term positive effects of decisions you make. If you’re that type of person, our future managing fleets of increasingly-powerful AIs as full time managers will be amazing.
Fireside True Story™ Time: One instance where I knew my job had gotten too meta happened while leading the Facebook London engineering office, which had grown from twelve original engineers into five hundred.
I found myself on a video call sharing the results of a spreadsheet I recently created to forecast how many interns the London office should hire the coming summer, in consideration of the likely H1B visa grant percentages next summer, such that Menlo Park could have the right number of full-time college graduate hires it wanted in two years.
The rewards of my work had become too indirected for me when I realized the approach I espoused during the call would take a full two years to prove right or wrong. I understand and believe the work I had done was valuable; in fact, two years later, it did turn out I was right. But the gap between decision and result had become too distant for me to feel fulfilled.
For the rest of us, the qualitative difference of direct work vs. meta-work is painfully tangible and unfortunate. For instance, lunch microwaving used to be a time of idle contemplation, an opportunity to think about whether my current PR can be better implemented. Nowadays before heading off to the kitchenette, I feel the stress of needing to allocate tasks for all my AIs prior to disappearing for three minutes. “They’d otherwise be idle! All that potential just sitting around unused!”
In honesty, I don’t even use the bathroom these days before prompting several AIs with work while I’m gone 120 seconds. I’ve been tempted, even whilst writing this very sentence, to first switch windows to occupy several AIs with tasks while I’m here typing. I’ve seriously considered, though have so far resisted, never going to sleep before spawning a few AIs on speculative long-running tasks I could check on the following morning.
The experience of being a manager is qualitatively vastly different from being an individual contributor. Direct work isn’t better or worse necessarily than meta-work, depending on your disposition. But it’s important to recognize we’ve passed an event horizon where a choice between the two no longer exists.
Welcome to management.