Alright you have forced my hand. You people need to be removed from your computer and banned from the keyboards. We don’t even deserve our own self-imposed title of “software engineer” or “AI engineer”. All of you yapping your stupid opinions online as grains of wisdom have left you naked in the wind exposing yourselves.

Many of you are dishonest and are misleading the field. It’s September 2026 and the hot debate online is “Should you read the code?” So now since you all can’t seem to be honest, I will set the record straight.

Framing the Vibes

Now before I start getting sights on the target, I will be honest. The people I am about to use as examples are more popular than me and by modern standards are more “successful”. Regardless, I am a nobody but I’m at least an honest nobody.

This all got started with the term “vibe coding” coined by a PhD named Karpathy.

This guy needs no introduction and is an accomplished researcher. Honestly, this isn’t even the issue; this is just what got the snowball rolling. This idea of just letting go and trusting the vibes would escalate to where we are today.

Models today are much better and the open weight models are also very impressive and can be better options than the expensive frontier that honestly has lost its capability edge compared to a year ago. So now since the models and the harnesses around them have improved, the general developer and even those semi tech adjacent can get these models to produce code.

This has spun out of control and now people are questioning: should we even be reading the code? If agents can produce thousands of lines of code in minutes then reviewing it all is stated as impossible or that people are burdens and bottlenecks of “productivity”.

Coding is Solved? Coding was never the hard part?

Boris (creator of claude code) has many videos and tweets on this. Coding is solved is what he and many others made popular.

Coding is the easy part or Coding was never the job is something many people have been spouting out. Let’s think about the stupidity of this.

Coding is a large part of the ethos of this industry. We are not milking cows, brick laying, making art sculptures, or making medicine in a lab. We are on the stupid computer you idiots. We are in the dark with our Doritos and Mountain Dew banging on the keyboard. We make software, hardware, and other tech related things. Coding is a major component of this job.

Coding is hard. The lighting engine in Unreal didn’t just appear from low effort. You debugging a race condition for hours or days is not easy. Figuring out the best way to pack the bits, thread the algorithm, or write the GPU kernel is not easy. Guys, not everyone works on Shopify sites. Some of us have to solve hard problems. The code is sometimes the very hard part.

This whole idea washes over all the work that people have done. The people saying this don’t code or have not worked on anything challenging. Don’t go around saying this crap. It makes you look dumb.

I understand the sentiment but it’s just not true. It is only true if you are doing something you have done before and know how to do. Yes these models can generate code faster than I can type. This is useful for many things but this idea that the job was never about coding or coding was the easy part is not true.

Imagine if an artist said it was never about the paint or brush it’s about how the painting looks… How are you going to make the painting without thinking about the paint and brush? I swear our industry deserves to be culled. Maybe these layoffs are justified and the real reason is as a whole we have lost our minds.

Software Factories Run by Stupid People

We are going lights out dark thirty on our super smart automated software factories where you only turn the lights on for a quick glance and then let the agents work and automate the entire programming/architecture of a product. This is the new buzz term. We are making labor camps for our agents to work in 24/7.

The only issue is these factories are glued together and on fire. The worst part is they are being built to create garbage. I can’t wait for the next Jira clone from your super impressive factory. Or maybe you are extra meta and your factory is building and maintaining other factories. You are a very smart human and making great choices in life.

All this work just to not read…

Surely we can be like other fields and design strict test frameworks so this can be possible? Let’s check up on how the man who wrote the book Clean Code is handling this.

Uncle Bob needs no introduction and has shaped many people’s opinions on programming. This is not a clean code rant so I will not be getting into that. Instead he has tried very hard to stop reading the code and use elaborate testing methods to control these agents.

Agents are chaos and have fatal flaws in their design that make it very hard to work with in code bases. They can be helpful and extremely frustrating. So Uncle Bob has claimed that if you write all the tests known to man then agents will obey and you can get good results.

Now I will be fair and say he admits this is a ton of work to even attempt this. We can’t read the code so we gotta let the tests try to make sure things are working because reading is hard. Also the AI writing tests for the code it wrote is many times not great.

If anyone could pull this off then it is Uncle Bob. If there is a man on this planet to make a software factory then he might be the best candidate to try. There are some positives to this. We should be trying to make good testing frameworks that give us confidence that the systems we build are working in the way we expect them to. Testing just tells me that the system is not breaking in the ways I would expect it to break.

Keep in mind many of these claims are from his missile command project. It’s even hard to get agents to make a clone of a well known game with many examples online in a way that you feel confident in it. Imagine a more complicated problem and env than “going dark” with wrapping agents in some factory loop. Good luck buddy.

The reason this doesn’t work in 2026 is because there is no amount of testing to stop the fatal flaw of token generation for models that do not have any world state or encoded world space to play in. LLMs are not world models (some debate this) at least not a mature world. Until LLMs have a world model and state space to see into branching futures then the software factories will always collapse.

There is no amount of tests that can prevent agents from going off the rails. When you remove yourself from the core (the code) and attempt to abstract yourself to a higher level then you are never going to have full control and have a product that reaches its potential. You will ship (if you even ship) a worse product over time.

The Copium is wearing off

You can only spew crap for so long. It’s fun to make these claims and sometimes when agent loops/graphs/factories work you think oh this is going to work all the time or I can get this to work 85%+ of the time. You get excited and think you cracked the nut but reality is still there in the code. The code and system are broken after months/weeks of lights-off development or even light review.

So how are the “coding is solved” people doing today? Look at this change.

This does not make any sense. It’s like the ultimate backtrack. When people have been saying coding is solved the implicit meaning of this is programming can be automated and has no need for people to do it anymore. Some gradient thought along those lines was the implication.

So now bugs are not solved but coding? Now the pivot is well coding is solved means typing the code… Well if getting the code onto the editor was the issue then coding has been solved since the modern keyboard was created. It’s a dishonest take. You know what you meant and now it’s time to backtrack because the truth is that coding is not solved.

Now let’s see how the “factory” is going.

Guess the factory is still fighting the master. You end up wrestling these agents to death. There is no lights-off development. You have to be on that factory floor beating these agents into submission. LLMs are not deterministic and will always go off the rails no matter how many guards you have hired.

But Muh Agent is like a Compiler

An LLM is like a compiler and allows people to move up the abstraction branch and just prompt and succeed… That is what people are saying.

Guys you work in the tech field. You program and yet you don’t understand or know how a compiler works. You are a drooling fool. Leave the industry now and find something else.

The reason this statement is crazy is because your compiler is designed to create a reliable build of your code. Compilers can be deterministic if all variables are controlled. Your LLM can write a great function in one prompt then switch libraries and conventions in the next prompt.

The argument is well no one reads assembly anymore. Well listen up idiots, you may not read assembly but the people working on your compilers sure do. People are still working full time jobs today making sure the compilers get better and produce the correct assembly. This paradigm doesn’t really exist for LLMs. You can’t force an LLM to make code always be the same. I feel like if I have to keep explaining this then I’ll go crazy.

Understanding Perspectives

In the case of Uncle Bob he is retired and does not have to participate in this meat grinder anymore. When he says these things take that into account. He sees agents as fun tools to play with and try new things with. I don’t think he is dishonest or trying to mislead anyone. I think his initial assumptions and measures of success were too short and small to apply to the industry.

He is a very successful developer and if he wants to spend his free time playing around and trying new things out with agents then go for it. He can do whatever he wants and there are some lessons we can take from his experiments. Just consider his perspective when reading his content.

Now with Boris I do not have a solid grasp on his true intentions. I don’t know him and so I will only state what I can see. He works at Anthropic. They benefit from you burning all the tokens possible. I think he just inflates the truth too much and has empowered this crazy cult of people who want to never read or understand code.

Now the bigger problem is all the other crazy voices of people who may not code at all or were never good at the job to begin with. These few years have taught me that many in this field hate programming and problem solving. If a CEO of a startup or company is yapping about their automation and loops just understand their position and companies.

Now Time for Reality

You get paid to understand the system you are building. You get paid to know where to improve the system. You get paid to try your best to make reliable systems. You get paid to solve problems at various layers. You need to read the code.

Being the “bottleneck” or “gatekeeper” is a good thing. You don’t want a bunch of idiotic agents dropping turds everywhere. No matter how many tests or guard rails I put up when I examine the code from the top models I still find many instances of bad/wrong code or code that might work now but will not later. You need to read the code. You cannot let these tools go dark for real products or things you are expected to be responsible for.

Everyone keeps yapping on their workflows but almost none of these people release anything or have a repo they want to show. When they do it gets ripped apart.

Now is the time to do the opposite of what others are saying. Embrace the chaos and dive deeper than ever before. Get excited to learn harder topics. Learn how to read the assembly, learn how to design different systems, learn to empower your team, and learn to rise above these idiots.

Conclusion

Programming with agents in 2026 requires even more knowledge than before in some ways. You have to understand how systems are built and interact. You have to think faster and harder about your decisions and be able to wrangle and run a few different implementations to see the best one. You have to be intimate and work hard to build anything with multiple agents.

There may be a time in the future where we don’t need to be as intimate and I think many people like Boris have done a good job in forcing people to see their vision. The vision is not reality in 2026. You need to wield these tools like a warrior and command them in the battlefield. You have to check their work and decisions. You can create wonderful new things as long as you put in the effort to learn and grow. Use agents as little or as much as you like but please READ THE CODE AND MAKE IT BETTER.