In August 1981, MTV launched by playing a song called "Video Killed the Radio Star." Forty-five years later, that song is still on rotation on FM stations. The Buggles wrote a number one hit about video killing radio, and proved themselves wrong from both directions. Radio didn't die. It adapted.
Developers have watched their tools get better every few years. It started with Stack Overflow. For fifteen years, that was the go-to place developers went when they got stuck, write a question, scan the answers, copy the highest-voted snippet, adapt it, ship it. An entire generation of engineers learned to code this way, and the running joke was that "senior developer" meant "better at Googling."
Then in late 2022, ChatGPT rewired the habit overnight. Why sift through five-year-old Stack Overflow threads when you could ask an LLM and get a custom-tailored answer in seconds? Stack Overflow's traffic cratered. The copy-paste loop didn't disappear, it just moved. Developers stopped pasting from a web page and started pasting from a chat window.
That was still a round trip, though. Prompt, read, paste, tweak. The next shift collapsed it. Tools like Cursor, Copilot, and Claude Code moved the model into the editor, onto the actual codebase, with full context of the files it was editing. The loop got tighter. The engineer went from asking for snippets to directing an agent that reads, writes, and tests on its own.
That's where the fight started. As soon as the agent could write most of the code, three camps formed around what the engineer should do.
In early 2025, Andrej Karpathy coined a phrase for the first camp: vibe coding. You describe what you want, the LLM writes it, you mostly don't read the code. If it works, ship it. If it breaks, prompt again. The name stuck because the behavior did.
Traditionalists pulled the other way. A decade of debates about code quality and maintainability collapsed into one question: if no one understood the code when it was written, who is going to understand it when it breaks at 2 AM? To them, vibe coding isn't a paradigm, it's technical debt with better marketing.
Between them is a quieter third camp: augmented programming. The engineer is still in the chair. They still read every diff, name the variables, own the architecture. But they aren't typing every character. They're directing tools, coding agents, refactoring assistants, test generators, that handle the parts of the work that don't require judgment, while the engineer focuses on the parts that do.
These three camps are making different claims about what software engineering is in 2026. Is the engineer a writer, a reviewer, or a conductor? Is code a craft, an artifact, or a byproduct?
AI is a tool, not a replacement
I started out in the traditionalist camp. To me, coding was a craft, an art. The IDE was a canvas. The final product was a work of art, shaped carefully over time, and I took pride in the quality of it. Vibe coding, in my eyes, was lipstick on a pig, it was still a pig underneath. I'd roll the dice on what the LLM returned, hoping it would work, and I kept finding myself doing more corrective work than writing.
Then I moved to augmented programming, not as a novelty, as my actual workflow. I produced more code, better documentation, and more thorough tests. When I was stuck debugging, I got unstuck faster. Refactoring a class that would have eaten an afternoon took minutes. The productivity jump was real.
But here's what I didn't expect: the craft didn't die. It evolved. I was no longer spending most of my energy on keystrokes and syntax, I was spending it on the parts that actually mattered. The architecture. The data model. Whether the abstraction made sense. The work I was proud of was the judgment, not the typing.
That reframing is what sold me on "AI is a tool, not a replacement." A good tool takes the mechanical parts and hands them back to you faster, so you can spend more time on the parts that require you. A saw is a tool. A lathe is a tool. A coding agent is a tool. None of them replace the carpenter, they let the carpenter build things they otherwise wouldn't have had time for.
Which is why I've come around on vibe coding too, though not as a production workflow. Vibe coding is a specification tool. When a client can prompt their way to a rough working prototype, something important happens: ambiguity disappears. They stop describing what they want in words and start showing you. By the time the real engineering team touches it, scope is pinned down in a way that a hundred requirements meetings couldn't have achieved. The prototype is disposable. What it produces, clarity, is not.
Three tools, three jobs. Traditional coding still matters when the craft itself is the value: a library, a protocol, a security-critical path. Augmented programming is the day job for most production work. Vibe coding earns its keep at the front of the process, where the question isn't how to build something but what to build at all.
None of those three is the future displacing the past. They are a set of tools, each good at something different, and the skill worth developing in 2026 is knowing which one to pick up when.
Conducting Mr. Roboto, AI at Lanovix
Every engagement is led by a senior engineer, and every senior engineer at Lanovix pairs their judgment with a full suite of AI tooling. We direct coding agents to write the scaffolding, generate the tests, draft the documentation, and run the refactors. When a bug shows up at a strange angle, we use AI to narrow the search before rolling up our sleeves. The engineer is still the one deciding what good looks like, the architecture, the data model, what belongs in the system and what doesn't. The agent does the typing. The engineer does the thinking.
The result our clients see is straightforward. Work that used to require a large team of project managers, senior engineers, and junior developers now gets delivered by a smaller, focused team in a fraction of the time. Budgets that used to fund typing now fund higher-leverage work, integration, infrastructure, the parts that actually move the business forward. Clients don't pay less because we cut corners. They pay less because we cut the parts of the work that no longer need a human to do them.
What clients don't give up is the senior engineer relationship, the clean architecture, or code that's maintainable six months after we're gone. You're still working with a principal who owns the outcome. The code you get is code that another engineer can read, extend, and trust. AI didn't replace that. It just let us produce more of it.
Radio didn't die. It adapted. The developers who adapted with it are the ones shipping software that actually lands, faster, cheaper, and without trade-offs in what it's like to work with them.
If you want to talk through what Lanovix's team can do for your business, get in touch.