Who the F*** Am I?

Hi, I'm Daniel Flores. A software developer with, I believe, seven years of professional experience.

It's been a wild ride — at least the past three years. I was one of the first to actually try GitHub Copilot during its preview, probably around 2021 or 2022. I was completely amazed by it, but it was definitely very, very rough around the edges.

Nobody knows me, though. I never intended to become a public figure or one of those guys who "knows where AI should go." That's not me, and that's okay. I've been watching this new paradigm evolve — and devolve — from the sidelines.

What I Think About LLMs and "AI"

I've been watching videos about this topic for years. Evolution simulators, learning algorithms, a model trained to play hide and seek — I was completely baffled when I realized that video came from OpenAI itself.

I'm not an expert in how to build models or LLMs. I have no PhD. I've just been doing my own thing for years, watching which tools get adopted and which ones suck. And I have to say this: the industry doesn't know what the hell is happening. Neither do I. Nobody knows, and that's what I hate the most.

LLMs are extremely useful. They can save you a lot of hours of work. But that's only half the story.

So What's the Issue?

Almost everyone — paid AI shills, mostly — is telling you: "YOU'RE GETTING BEHIND IF YOU DON'T USE THESE TOOLS!!"

They're trying to push the entire industry into a fear-of-missing-out state.

Let me share my personal experience about this completely inconsequential fear.

If you're already experienced enough — if you already know what an LLM is and can prompt it to do or refactor something — you're not missing out. That's all there is to it.

I'm going to explain what I mean. The exact same issues I had with the old GitHub Copilot — I don't even know what model it ran, probably a customized GPT — are still true today with the latest frontier models.

They all hallucinate. They all seem to kind of understand what they're doing until they don't. And if there's no human in the loop to catch the mistake, it compounds across every agent in the chain.

This is old news for experienced developers who've already tried every agentic loop, every CLI, every local and API-hosted model. They all share the same fundamental behavior. They're not exactly "mistakes" — the LLM is just showing you the most probable next token — but you get the point.

But You Should Still Be Scared

I'm scared — but not of missing out on LLMs or technology. I already know the limits of these systems, or at least I believe I do. I'm constantly learning how to build more efficient tools with them. Right now I'm using Xiaomi's MiMo v2.5 model. For your standard vibe-coder, that model is absolute trash. For a seasoned developer, it's fantastic and extremely cheap.

What I'm scared about is the future of the industry: the jobs, the way we sell our knowledge.

I'm currently taking a new stance. My CV applications explicitly state that my LLM-assisted approach involves reviewing 100% of the generated code. But that's not sexy. That's not standard. And I'm sure that's why nobody has been calling me. That's partly my fault — I could probably sell my skills with better wording — but that's not the real point here.

The standard advice everyone gives you is: "LEARN TO USE THESE SKILLS, TOOLS, HARNESS THE NEW THING." Until the next thing is released. Oh, but they won't pay your bills while you spend your own time and money testing it. Of course they won't.

I'm not scared of learning something new or trying out the next model, because most of them are just slightly better at being autonomous or solving a badly explained problem.

What I'm scared of is that the industry wants to move fast toward the next model, and the next, and the next — and it doesn't matter how expensive creating a new model is. They just won't slow down.

New models are being released every day. But not a single standard accompanies them. Vibe-coding is already reaching companies, and I really hope no insane organization lets unreviewed vibe-coded software touch a pacemaker or an embedded control system.

The Real Problem: Drift

I know — and a lot of people know — that these token-gambling machines are imperfect. But they're useful. However, they're flawed.

I've already tried a lot of experimental workflows and approaches. Spec-driven development — I've tried it. It works, somewhat. But the moment the model decides to drift, it will drift.

It doesn't matter how well-written the spec is. The model will drift when it wants to. And if you don't review the source code itself — even at the most critical step, at the final output — it can still decide to drift. This drift doesn't just mean writing a small bug or the wrong logic. It can also mean the model decides to invoke a sub-agent to do an extremely comprehensive and wasteful research task about why a single function should exist. This is just one example.

When a model drifts, the impact can range from negligible to fatal.

Humans drift too. We're lazy. And honestly, most LLMs are better than me at coding on a purely technical level, most of the time. But I have an ego — I care about the work. The LLM doesn't. It's a gambling machine. It can't decide for itself.

We can decide.

What I Truly Believe

I'm not saying 100% of all code must be reviewed. But I firmly believe:

  1. All unit tests, integration tests, and end-to-end tests should be 100% human-reviewed. Even if the tests are hard to understand, make the agent refactor them until your brain can actually ingest them. If you can't understand the test, you can't trust the test.

  2. The most critical parts of any system must be reviewed 100% of the time — every single line. I don't want to lose my hope in humanity if a block of vibe-coded slop ever touches a critical system and kills people. (As of right now, I haven't investigated if that has happened, but I believe it is very possible it has already happened).

Conclusion: You're Not Falling Behind

Let me say this clearly: you're not getting behind.

The industry is shifting rapidly, I'll give you that. But nobody — and I mean nobody — knows what the hell they're doing. Not me, not the companies building the models, and definitely not the guys selling you courses. They don't know shit either. This is all too new.

Nobody is realistically using LLMs with 99% accuracy, efficacy, and efficiency. Not a single person on this planet. Anyone who tells you otherwise is either lying to sell you something or lying to themselves.

Every few months, someone will sell you the latest course, the latest workflow, the latest "one weird trick to 10x your output." And when it becomes "obsolete," you'll think it's because something better came along. But that's not what happened. It became obsolete because it never worked in the first place.

The model changed. The API changed. The hype moved on. And you're left holding a course about a tool that doesn't exist anymore.

So here's my advice: stop chasing. Learn the fundamentals. Understand what these tools actually do and what they actually can't do. Build your own judgment. That skill — knowing when to trust the output and when to throw it away — doesn't become obsolete. Ever.

The hype cycle will keep spinning. Let it. You're not behind.

You can do yourself a favour and start using Pi Coding Agent too, it's a good opensource tool. OR any other opensource agentic tool of your liking, opencode is a good bet too. And just do that, start experimenting with it, no need to download skills, no need for any of that... just try to investigate what the raw LLM is capable of doing, and expand from there.

Note: Btw you might have noticed, yeah, I used a model to correct this post, MiMo v2.5. I don't know if it truly got corrected, but I believe it did a good result, it preserved most of my personal touch here. It even added its favourite em-dashes all over my original draft. Whatever, I'll let it pass today.