I started my career as a software engineer, but hadn’t written production code since I was at Uber in 2018 (and even by then it was mostly just ML models).
But over the last month I’ve shipped ~190,000 lines of code across 11 internal repos with the help of Hermes, including building a fleet of internal agentic tools from scratch.
And I’ve been able to cut $425,724 in annual expenses as a result (with only a fraction being put back into AI costs) while simultaneously dramatically increasing shipping velocity on the team.
I’ll do a breakdown on:
What those costs were
The system I’ve built to make it possible without sacrificing quality
What’s next (cutting and efficiencies are good, but growth is better)
But the question that was on my mind consistently over the last month has been what the real job of humans actually is now.
Or, without getting too grandiose, what existing and new roles are still best suited for humans?
This week I’m sharing what the 5 I landed on are.
The 5 Jobs of Humans
Of the $425,724 that I cut, roughly 59% ($250k) was payroll and contractor expenses.
But I’m not a solopreneur and likely can’t currently see a world where that path makes sense for me.
Here are the jobs I think are uniquely still best done by humans:
Ambassadors
Remember when Klarna replaced their customer support team with agents, only to fully pivot back to humans?
People thought that was a critique of AI but, really, it was just that Klarna realized the last place people want to talk to an AI is when they’re frustrated and dealing with an issue.
This principle applies across all customer facing roles. Your brand needs ambassadors, who are human, when interacting with other people.
Hell on earth is:
AI interviewers
AI customer service reps
AI sales reps
AI partnerships people
So here’s a quick thought experiment:
Can you imagine trying to successfully raise capital for your company if you had an AI do it?
Boardy nonwithstanding, the answer is almost certainly no.
And it makes sense. If AI is making it easier to do a lot of things, people want to see (and work with) people who value human interaction and relationship building.
Context Farmers
Someone needs to build and maintain the system with a sense of product judgement.
I no longer call these engineers but, rather, context farmers (s/o to my former co-founder for writing about this a few months ago).
The goal is to feed the right information, formatted in the right way, into the right system in the right order and at the right time.
They’re “farming” for the right information and then nurturing and growing the system.
This has basically become my role within Rye Valley (my holdco) but, in theory, you could have different parts of this done by either a junior or senior person.
Senior folks would be needed for the architecture of the system, but as far as sourcing the information, you could hire subject matter experts who don’t know too much about your company or even startups in general, and just have a knack for research or some domain-specific knowledge.
So far I haven’t delegated this mostly because my speed hasn’t been a blocker to anything yet.
Captains
If farmers build the farm, captains steer the ship.
Captains are senior people who deeply understand everything in their domain, the quality bar, the edge cases, and second-order effects of the decisions they make.
Their job is to make sure the system is doing what it’s supposed to do for the end user. They evaluate outputs, tweak the constraints, and know what “good” looks like when agents start going sideways.
The more capable systems become, the more important having the right captains is.
They know more about their respective verticals than I do, and ensure we’re staying on top of shifts in their industries.
I could see a future where you have junior captains (first mates? lol) who get partial responsibilities delegated to them, which is yet another datapoint supporting the idea that AI will change jobs rather than remove the need for them.
Referees
The toughest part of these recent changes was reducing the hours of some key people on my team.
But instead of cutting them entirely, I realized pretty quickly that if agents are creating outputs then we’ll need someone to have the authority to “stop play” when something is off or before those outputs go anywhere important.
Basically, a human QA layer that isn’t limited to code but rather is across the entire business.
This is taste, applied in a polishing and critical context rather than a steering context.
It doesn’t mean everything needs to be reviewed, but in some areas it’s crucial:
Subjective and/or brand-related work where “correct” is fuzzy
Work without a clear “definition of done”
Investigating edge cases the system hasn’t been taught about yet
High-stakes outputs where the cost of a miss is large
Calibrating whether the current quality bar is still the right one
At Rye Valley, this is done by my chief of staff. He knows my preferences, and our business, better than anyone and he has a high attention to detail.
Stewards
The best team members can be context farmers, captains, and referees all in one.
That’s the modern definition of seniority.
But there’s one last type of role you probably want to keep separate:
Capital allocators (aka stewards).
These are the people who determine budgets for compute, and other irreversible decisions around spend that can impact the overall health of the business.
In most cases, this will be the founder for a long time (maybe forever) since it involves decisions around acceptable burn rate, fundraising, and when to shut down projects entirely.