AI Is About to Change Coding Forever in 2026 – “Software Engineering Is Done”

Coding is just writing text. All "AI" does is add 1 token at a time to a text it is given. It doesn't think. Yes, it is useful, but it shouldn't be overhyped.

Credentials, job titles, brand names, and buzzwords are irrelevant. Output quality is the filter. If you cannot ship reliable systems, reason under uncertainty, and fix your own failures, you are not employable. 2026 does not reward passengers. It rewards people who can build, operate, and debug real systems end-to-end. Everything below assumes that baseline.

First principle: fundamentals dominate. You must be fluent in one systems language (C++, Rust, or Go) and one high-level language used for orchestration and data (Python is the default). You must understand memory, concurrency, I/O, networking, and failure modes at a level where you can explain performance regressions without guessing. If you do not understand how Linux actually schedules work, how TCP behaves under loss, or why your service falls over at scale, tooling will not save you.

Second: production engineering, not tutorials. Containers, but with an understanding of cgroups, namespaces, and seccomp rather than just Dockerfiles. Kubernetes, but with actual operational knowledge: scheduling, resource pressure, evictions, networking, ingress, and failure recovery. Infrastructure as code that survives audits and disasters. Observability that lets you answer "what broke and why" in minutes, not hours. Logging, metrics, tracing, and alerting wired to real decisions.

Third: data and state. SQL is non-negotiable. Not ORMs actual query planning, indexing strategy, locking, and consistency trade-offs. At least one distributed data system understood deeply enough to know when not to use it. You must be able to reason about state, migrations, backfills, and irreversible mistakes. If you treat databases as magical storage buckets, you will be filtered out quickly.

Fourth: AI as an accelerant, not a crutch. You are expected to use models to increase throughput, not to outsource thinking. That means knowing how to evaluate outputs, constrain tools, wire models into real pipelines, and detect when they are wrong. Model APIs, embeddings, retrieval, and basic inference economics matter. Prompt spam does not. Anyone who cannot function without an assistant is a liability.

Fifth: security and reliability. Threat modeling, basic cryptography literacy, secrets handling, access control, and least privilege are table stakes. So are backups, disaster recovery, and knowing how systems fail under stress. If you cannot design something assuming it will be attacked and will break, you are building toys.

Final filter: evidence. Public repos that show sustained, boring, correct work. Systems that run, not demos that impress for five minutes. Clear written explanations of trade-offs and post-mortems. If your output looks like glued-together snippets, you will not make it. If it looks like something people can rely on, you will.

I am ready for it, but you know what? Let's as programmers start using ooen source models feed them with our codes and questions. So our efforts make an open source god model before becoming "obsolete"

And I was thinking about getting into freelance web design😂

Take me back to the 2000s, present is bleak and future is dystopian

These executives are full of shit. There is no model that can work for 30 minutes autonomously and deliver desired high-quality output. They can't even do 2 minutes in practice without handholding, but they will claim whatever crap to make people buy their product.

Machine code produces whatever is in the high level language. It doesn’t know whether code is buggy or not. It produces whatever the developer instructs it to do. The reliability is still on the developer who instructed the computer to do. Similarly if AI produces buggy code who is going to verify it? Is AI will run its own code or humans will run it to find bugs or reliability. Has any agentic software went to production if so do we have any video on production based agentic software? I can’t trust AI code. It’s a dangerous statement

People should halt buying or renting or investing on these defective products. They advertised AI won't make mistakes, very secure and very cheap, yet on reality it makes lots of mistakes, with multiple security holes and very expensive to run (expensive chips + electricity hungry). They are doing so much fraud, false advertisement, defective products and massive engineering malpractices.

Oh yes, the whole world will run on "vibe-coded" software in 2026 🙄 Better get used to bizarre hallucinated settings in your operating systems and rebooting your devices 20 times per day due to memory leaks and infinite loops.

No, it wipes out human life. Humanity is fuzked. RIP humanity

It is wild to think that the future of development is just being a really good architect. You use the heavy AI models to build the complex systems, but for the simple stuff, you do not even need them. For a basic project, I have sometimes just used a drag-and-drop tool like Tiiny Host to avoid over-engineering, but for real software, these AI agents are going to be doing 90% of the work. It is a completely new world.

Didn't Anthropic's CEO predict that software engineering would be over by end of 2025?

What happened? 😂 these people are professional conmen and the AI bros just eat it all up.

The volume of your last few videos is REALY quiet I don't use a head set and I had to crank up my volume in order to hear you.

Opus 4.5 in github copilot is amazing. I only wished it googled more often hard solutions

Big companies won't implement AI tools for a while… security reasons..