AI Coding Black Hole

Lately I stopped trying to resist the pull of the AI coding black hole. By that I mean I do not code by hand anymore. I don’t even bother changing variable names, even though I have some meh-to-okay vim skills (shout out to ThePrimeagen for his vim series, but these skills are going to get obsolete pretty quickly¹). The news we hear about engineers at company after company only using AI to code is actually true.

When I first thought about this black hole analogy, I was going to write something like “if you have large mass, like Linus Torvalds, you may be able to resist the pull of AI coding.” We know that wouldn’t have aged well. Yes, it wasn’t really production code, but it all started that way. Speaking of ThePrimeagen, I believe he is a very good programmer (relatively big mass) and genuine in the content he creates, in the sense that his primary drive is not generating money (his trajectory not pointing directly towards the black hole). But he too has been leaning into AI coding lately. Seems like we are all headed in the same direction.

But why did we resist, or still resist, AI coding? A lot of it comes down to identity. We define our abilities by the quality of the code we can produce. When a language model generates mediocre code on the first, second, or maybe even the fifth try, and we still use it in our sacred codebase because it was faster than writing it ourselves, it reinforces the feeling that relying on it diminishes us. At our core, we associate manual fluency with competence, so letting AI code for us feels like waving the white flag.

Then there is the skill decay. We have learned and forgotten many skills and tools before, but this feels different. As developers, we build confidence around mastery of our primary language, its syntax and edge cases. Now we forget the basics of how to wield our primary weapon. If we are no longer the ones who remember every corner of the language, what are we? Our identity as developers is slipping away and being replaced with something else, and nobody knows what that something else is yet. I could mention other reasons too, but they mostly stem from identity, just reframed.

Getting past this is hard, especially for ICs. If you became a manager at some point, you already went through this whole dance, despite how hard it was. Yet, you got the promotion (kudos to you by the way), which made it easier to accept the new reality. But an IC opted out of the manager role by choice, and now they are being forced to become the managers of AI agents with no apparent rewards. The people who are most excited and bullish about AI coding tend not to define themselves as developers in the first place.

Now, all of this has been happening right under our noses, and it was not a single decision but mostly a drift. We used GitHub Copilot with Codex (the model) back in 2021 for code completion, and we remember how insane it felt at the time, something everyone takes for granted now. Then came ChatGPT in 2022 with the back-and-forth copy pasting, having it write tons of lines of docstrings and unit tests, finally doing the boring parts of the job, but hey not the actual job (no offense to QA engineers). Suddenly, in 2024 it slipped into our actual code logic with Cursor². Finally in 2025, Antigravity, Codex (the agent), and Claude Code, drove the final nails in the coffin.

So, depending on your gravity and trajectory, either you have passed the event horizon at one of these milestones or you are about to. Like most of us, I think I had median developer gravity and a moderate trajectory somewhat pointing towards the black hole, and got sucked in within the last month. Linus Torvalds has considerable mass, and kernel development is a different kind of work (on the other hand, the Linux kernel is already making some room for AI-assisted contributions), but I believe he too will eventually get sucked in.

The weird part is how distorted the perception becomes. From inside the event horizon, we feel the productivity gains. Output goes up, things ship faster, we feel like 10x developers. But from the outside, time has moved on. People have already gotten used to this speed. What feels like a breakthrough to us looks like the new baseline to everyone else. When 10x is the norm from the outside, devs at 2x or even 5x start to look bad, and we wonder if we will make the cut.

To put some numbers on it: it took me a couple of hours on a Sunday afternoon to launch this website from scratch. It is not a page on Netlify, but a properly set up static site: Hugo on S3 behind CloudFront, all infrastructure in Terraform with remote state, OIDC federation between GitHub Actions and AWS so there are zero stored credentials, CloudFront access logs queryable through Athena, and CI/CD that runs terraform plan on PR and deploys on release bla bla bla. Nothing too fancy though, just done rightish. Writing this blog post took me twice as long.

What about code complexity, maintainability, security, all the other -ities? I initially thought it would take a while to solve these. But right now we can throw some decent SDLC workflows with SDD and TDD in the mix and we are good to go. And even that level of involvement may not last long. Planning mode, hooks, skills, subagents, agent teams with PR agents, security agents, plugins, all of it will be automated. With all these agents, might as well think we are in the matrix.

Among those -ities though, security is the one that gets talked about the most. It must be both a dream and a nightmare to be working in cybersecurity right now. It used to be the case that only managers did not prioritize security, now even half of engineers let it go completely. Everybody is leaving their newly AI-made doors unlocked, at this point you can find a free API key for most services with a manual GitHub search. OpenClaw followed the infamous PlayStation Network playbook, dialed the security down to almost zero, and people still went crazy for the product. On the other hand, cybersecurity devs must feel the fear too, with all those security and pentesting agents coming for their jobs. OpenAI’s Aardvark, and now Anthropic’s Claude Code Security. Cybersecurity stocks lost 10% in 24 hours after that last one.

Hard to draw conclusions when we are still in the middle of it all. It also feels like an impossible task to keep up with everything happening in this new world. Andrej Karpathy coined the term “Vibe Coding” but even he feels behind. So, try to remember that it is okay to feel overwhelmed.

When the dust settles, we will all need to find our new identities. Maybe you will become a product person, or a forward-deployed engineer but for AI agent clients, or go offline and go blue collar. Maybe you will be lucky enough to still practice the lost art of manual coding, at a distant star orbiting the giant black hole of AI coding.