Starting something new was never this cheap.
That’s where I landed after about a year of agentic coding.
How it used to be
Until agentic coding arrived, there was a natural balance between what software needed to exist and what software shouldn’t - because the price to solve a specific problem with software would’ve been way too high.
Remember, back in the days you needed to find a human that even understands how to write code, and then also offers the specialization you’re looking for - web, mobile or desktop. The more specialized your need, the more expensive it was usually.
Before, ignoring code generation, ~every line that existed was there because another human - for whatever motivational reason, money, passion, fun etc. - decided it’s worth investing their time into writing that line of code. Trying to build or solve a problem they cared to solve.
It was slow, not necessarily better - software always was buggy because we all are imperfect beings. Computers always did, if they work correctly, only exactly what a human once told them to do, if you so will.
Just start
We all heard that the most difficult part of achieving something, building something, doing something, is to just start.
Before, when you had an idea and broke down in your head what it would take, you always made trade-offs: “OK well, this would take me at least a couple of days after work… so I don’t start.”
That trade-off in your head killed more ideas than any technical limitation ever did.
btw, this is something I constantly did when I still was more junior as an engineer. I built so many random small games or prototypes of mechanics, products, experiments etc. Over time I got better at predicting already what the outcome of trying something might be.
Which sounds like a skill, but it also meant I started fewer things.
Today that skill is worth a lot less. Trying became cheaper than predicting.
And start we did
Now there is a never-seen-before wave of throwaway software flooding the internet.
I’m generalizing, but I would argue the amount of new software that solves novel problems, in relation to the amount of code that got written that a human didn’t even read, is probably several orders of magnitude apart now.
And we all are part of that wave.
I’m noticing, looking at the graveyard of published projects on my GitHub profile - and even more unpublished ones! - that at one point I just stop caring about them. Way quicker than it was in the past. Mostly because I didn’t build any attachment to the code at all.
I prompted an agent and fixed a problem I thought about or just directly had in front of me. Sometimes this resulted in helpful extensions for my agent harness pi-config - and sometimes they even turned out helpful to others as well, see glimpse or pi-interactive-subagents.
But even for those libraries - which have more stars than anything I ever did in Open Source before - I really don’t care about them.
Code really became an artifact of my intent, more so than anything else.
Does it actually work?
To me, syntax, coding styles, languages, tabs/spaces - all those things really never mattered much. What I always mostly cared about, above everything else, is: “Does it work?”
And that’s the one question agents haven’t made cheap.
Don’t get me wrong, AI obviously makes certain tasks and aspects of software engineering way faster. There’s a whole new class of automations we can build now, with an LLM as the glue between things that are hard to codify in discrete code - the LLM just takes over that hassle. A whole host of new problems became solvable, also for production things, and it can make you a lot faster.
But everything is so easy now that sometimes you look a bit too far into the abyss. You start something that would’ve been a multi-week project before, you’re 80% there after one afternoon - and then you discover the outcome is not fully there yet. It still takes a lot of effort to bring it to the place where it actually does what you want.
And building a product that people actually care about? That’s still equally difficult as it was before.
Starting became extremely cheap. Making it actually work did not.
Where this lands
So I think it’s clear that we all are experiencing a shift in how software - and everything around it - is changing, in ways no one really knows yet where it lands.
Just read the tech news, we do things now that people thought are absurd 2 years ago.
So go start something. You might just learn or find something new.