Set and forget

Every leap in computing has handed people back a piece of their time by taking one job off their hands. The job still waiting to go is the one that keeps software alive after it is written.

An essay from LILY Labs

The history of computing can be told as a list of jobs that stopped being jobs.

In 1957, John Backus and his team at IBM released FORTRAN, and programmers stopped writing most of their instructions in the language of the machine. They described the calculation, and a compiler worked out the rest. Around 1960, John McCarthy, at MIT, built automatic memory management into Lisp. A whole category of bookkeeping, deciding when each piece of memory could be given back, moved from the programmer to the system. In 2006, Amazon put storage on sale at fifteen cents a gigabyte-month. Now a company could stop buying and managing their own hard drives and keep access to their data anyway.

Each time, people worried that something important was being lost. Each time, the people freed from the job did not sit idle. They built more interesting things on top.

None of these transitions was clean. Compilers produced slow code for years. Garbage collection brought pauses of its own. The cloud swapped buying machines for configuring them. Every handover created some new work alongside the work it removed. But the direction stayed steady. A job that used to fill a day became a thing the system took over automatically.

Writing code has never been faster. Tools that help with it, many of them now built on AI, can produce in an afternoon what used to take a week. The odd thing is how little that has changed the rest of the week.

Writing out the application logic is only one step. After it is written, someone has to package it, choose a machine for it, decide how many copies to run, set rules for when to add more, watch it, patch what sits beneath it, and answer the phone when any of that goes wrong. None of those tasks got meaningfully faster while the output of code exploded.

Komodor, a vendor for Kubernetes troubleshooting tools, reported in its 2025 Enterprise Kubernetes Report that operations teams spend more than 60 percent of their time troubleshooting. Even if discounted by a large margin, that is still a sizeable number. The bottleneck has moved. It used to be in the creation of software. Increasingly it lies in keeping that same software running.

The one-person organization

AI writes the code. Payments, email and sign-in arrive as services. Set and forget takes the servers off the list. Stack those together and one person can do what took a small team a few years ago: build a product, ship it, and keep it running.

And it is leaving the screen. CNC mills, 3D printers and laser cutters now fit in a spare room, and AI can walk a beginner from the first sketch to the finished part. Someone tired of a bike part that keeps breaking can design a better one over a weekend, print three prototypes, mill the final version in aluminium and be selling it a month later. The shop, the order queue and the shipping emails are just software. Set it once, and the attention goes back to the product.

Just open YouTube, the creator economy has exploded with people building things on their own that used to need a workshop and a team: electric conversions in home garages, hand-built instruments, hardware products shipped worldwide, and courses teaching everyone else to do the same. That flood of content is a public record of the new tools at work.

A product for a few hundred customers becomes a real business when all of them can be operated from a single hand. An idea can be tested within a week, not after a funding round. And the most valuable skills shift from depth to range: knowing enough about design, materials, code and customers to connect them, and to spot when the AI is wrong. Deciding what to build is still the hard part. Everything around it is getting lighter.

The freed hours can go in two directions:

The first is to do more. One person directs a small organization of agents, each carrying part of the work: E.g. one writes code, one answers customers, one watches the stock, one drafts the next product page. The person in the middle decides, checks and connects as the central orchestrating node in this network of agents.

The second is quite the opposite: Much of a working day is not work at all. It is clutter: searching five internal tools for one document, filling in the same fields on a contract for the tenth time, photographing a flat and clicking through every menu it takes to list it for rent. Each step is a tiny decision: the click of a button, locating the next textbox to fill out, remembering which note taking tool holds a specific piece of information, etc., etc. Together they can eat hours in front of a screen. That is exactly the kind of work agents are good at. The human in the loop simply needs to track the outputs to deliver. A lot of the workflows, machines can now take over just as well as an actual personal assistant, freeing up the users time.

Most people will mix the two. For decades, software competed for attention: more features, more dashboards, more reasons to log in. The tools that win next may be the ones that ask for the least of it. A product that does its job while nobody is looking stops being one more thing to manage and becomes what infrastructure was always supposed to be: the part underneath that lets everything else happen.

Back in the forest

It would be easy to end on productivity. Ship faster, do more, fill the freed hours with more of the same.

That is not quite the point. The people who stopped writing machine code did not spend the time writing more machine code. They spent it on problems that had been out of reach.

There is a catch. Agents shorten the loop between decisions. A change that once took a week to build now comes back in an hour and asks for the next call straight away. The manual work that used to sit between choices, and quietly gave the mind a rest, is disappearing. What is left is a day made almost entirely of decisions, and decision fatigue grows with it.

The same shift also loosens where work happens. You can walk through a park with a pair of smart glasses, talking an agent through a rewrite of your backend. Some of the freed time should go to nothing at all. Time, to replenish the cognitive resources spent. Deploy with us and rest easy, we will take care of the rest!

You build. We carry.

Start for free. Read the docs. Talk to us about migrating a workload.