Korva is a scroll of the internet’s best ideas.

The usage bar

Today the math works out only if a product will be used something like a million times, or maybe 10,000. Engineers are expensive, salaries have to be paid, and the software has to create enough value to cover all of that. YouTube clears the bar easily: far more hours have gone into using it than into building it.

Work below the bar

Plenty of work sits below the bar. Picture a morning spent looking through 15 LinkedIn profiles for one detail, filling out a stack of forms, or building a spreadsheet from research. Software could do every one of those jobs, but nobody hires a team to write a program that runs once, so a person does the work by hand.

Code nobody sees

With an agent, the person gives instructions and the agent works out the rest behind the scenes: it writes the code, drops in a script to automate one step, and moves to the next. The person never has to look at any of it. Software stops being a product built for millions and becomes a way to control a computer for a single job.

Why is custom software so expensive to build?

Software engineering is skilled labor, and writing the code is only part of the cost. Every program also has to be tested, secured, documented, and maintained as the systems around it change. Studies of software projects have long found that maintenance accounts for more of a program's lifetime cost than the original build.

How does an agent run a script mid-task?

An agent is a language model wrapped in a loop that can call tools. The model writes out a command or a block of code, a surrounding program executes it in a sandboxed environment, and the output is fed back to the model as text. The model reads that result and decides on the next step. Coding agents run this cycle many times within a single task.

How did spreadsheets let non-programmers skip engineers?

Before spreadsheets, a one-off calculation for a business usually meant asking a programmer or running a batch job on a mainframe. VisiCalc, released in 1979 for the Apple II, let accountants and analysts build their own models on a grid. Many historians of computing credit it with pushing the Apple II into offices.

More from David Senra

Tech & AIDevin's first task: set up MongoDBDevin's first real task was setting up MongoDB, and when it worked, one of its builders was too wired to sleep.
Enjoyed it?