You all have destroyed this industry. You are a token addict and need to be admitted into a mental institution. You woke up and started prompting with your 3-5 different AI subscriptions. It feels good, like a drug straight to the veins. You begin making all these things that would have taken you months to research and create. You just made a landing page in React, and you don’t even know what the DOM is. It feels powerful, all this “productivity,” and for only $20 a month. You feel the voice in your head. You are unstoppable. You have ascended and become a 10x engineer.

Until you’re five weeks in and your project keeps breaking in small ways every time you merge to main. Next thing you know, your database has been blown away, and you can barely maintain 70% uptime.

AGI/ASI Is Far Away from Your Agent

Unlike Richard Dawkins (who thinks we are all just matter and chemical reactions, and therefore Claude, or “Claudia” as he called it, is conscious and intelligent), we know that the LLM has no real human reasoning in it. Many people are mad that I said that and have raged out of this article by now. For those who still have some brain cells left, we will continue.

No one can define AGI/ASI, so it’s mainly a pie-in-the-sky metric, but one thing is certain: your agent does not meet that goal. You are more capable and can create better code and systems than multiple parallel agents left to their own madness. The more agents you run in parallel with no guidance, the worse the output is. The kicker is that even when whatever feature or workflow seems correct, there are thousands of small cracks in the system. You can’t trust the agents to write all these tests and just assume things are working. They are going to break the system eventually.

Look at just this month:

  • GitHub Copilot/Actions (May 2026): GitHub reported Actions hosted runner failures and delays in East US; Copilot Code Review was also impacted.
  • OpenAI Codex (May 2026): Codex Cloud Tasks degraded, alongside nearby incidents for Responses API errors, Codex transcription failures, and model-specific error spikes.
  • Cursor Cloud Agents (May 2026): Cursor reported service degradation across Cloud Agents, CLI, and IDE surfaces, with Anthropic-backed agent surfaces affected.
  • Claude Code (May 2026): An infrastructure change altered Anthropic’s outbound GitHub IPs, causing Claude Code remote sessions, GitHub Enterprise plugin syncing, and Claude Security to fail for allowlisted orgs.
  • Netlify Agent Runners (May 2026): Netlify reported Agent Runner failures affecting builds for newly-created projects, plus separate elevated errors and latency on hosted sites that same week.

These models are not superintelligent. These models have a large issue. They are only as good as you can make them, and you are not that good. You cannot move at the speed of sound and approve a 10k+ PR. It’s nuts. We are not meant to be doing that, and the agents, as of today, are nowhere near good enough to pump out that much code and have it work in production. Guess what all you AI hype idiots need to do: slow down, use agents on a smaller scale, and read all the code.

That’s right, read the code, moron. You need to understand what the system is doing, and that means you get to gear up for some gnarly code reviews. It’s part of your job to solve problems and provide value. Your 10k slop PR is garbage. You are the gatekeeper and bottleneck. Sure, maybe your crappy tool or basic web page can be glanced at, but the real system running in production needs you to be present and active.

Even internal tools need reviewing. Everything has to scan your eyeballs to some degree. If you are doing your job well, you realize that a lot of the initial speedup gains get crushed by the amount of change and review you have to do in order to solve the actual issue you are working on. You don’t get to hit the easy button for free. There is a cost.

All Software Sucks Now

I was promised 10x engineers, but instead I got 10x the bugs. Look at the state of all these “AI native” companies now. Discord is down as I’m typing. Coinbase has been down for days. Cloudflare went down a few moments ago… I mean, what on this earth is happening? Software is worse than it has ever been. All my apps are breaking in weird ways, or they require restarting or cache clears more often. Things were held together with sticks and glue before agents, and now I guess just the sticks are left.

You were average at your job or slightly above (nothing wrong with that; it’s an average for a reason), and your solutions were on a spectrum of good and bad. With access to an agent, you did not get better. You are probably making worse solutions than before. You still have to work very hard to get a good solution, which is why you have to slow down and focus on one area. Agent swarms are largely useless, and you need to get your hands dirty and write the code yourself or prompt the model like a slot machine to morph it into what it needs to become.

Productivity is All a Lie

These companies claim all these productivity gains, and all I see is downed apps and broken updates. Let’s focus on you for a minute. You get a slight boost for the first 50-70%, but then you spend a lot of time redoing that same work and a large amount of time on the last 30%. In many cases, you were better off not prompting and fixing it yourself.

These are just tools, and I hate how companies are expecting their devs to run 100 agents and produce at breakneck speeds. It doesn’t work. The quality of everything suffers. In some cases, you can really move fast if you know the system and the scope is small. If you know how to make systems and you know how to make good solutions, then you can get some speedup, but you still are not 10x.

In fact, I’m having to work harder than ever to get all this crap done and review all these garbage outputs. It’s so exhausting trying to wrangle these agents.

You Won’t Be Able to Afford the Tokens

Another issue that deserves its own article is the cost of tokens. Right now, we are seeing the end of these subscription models and the beginning of paying API rates. These API rates are all going to go up. Right now, these companies are struggling with compute, resources, and hosting. Another issue is that in the coming years, these investors are going to demand some of their money back. You can’t rely on prompting the best models for everything. You simply cannot afford it.

Conclusion: Strive to Learn and Get Good

Here’s the deal: you need to put the vibes away and actually invest in skills. You can become a great developer or engineer, but you still have to grind at it. You need to study systems and how to create them. You need to know the tradeoffs between technologies. You need to understand how to code. No one is good enough to turn their brain off and keep prompting. You will not provide lasting solutions. This industry is in a mess right now, and I hope we can all survive it. Many developers and leaders have lost the plot.

The pendulum will swing back. This cannot work forever. Agents are great tools when you know how to use them, but let’s use our brains and be intelligent about them. Please do not give up on becoming great. Do not lose hope. If you love this field, then continue to grow. We will get past this together.