Opinion

Some Developers Are Quitting Vibe Coding: What No-Code Builders Can Learn

Some developers are walking away from AI-generated code and returning to hand-written architecture. What that backlash means for no-code builders.

Some Developers Are Quitting Vibe Coding: What No-Code Builders Can Learn

In May 2026, a Hacker News post titled "I'm going back to writing code by hand" racked up 1,038 points. The author wasn't a Luddite. They'd spent seven months letting Claude build a Kubernetes dashboard called k10s for them, shipping feature after feature and feeling productive. Then they sat down and actually read the code they'd been shipping.

model.go was 1,690 lines. Its Update() method was a 500-line function with 110 switch/case branches. A single struct held the UI widgets, the Kubernetes client, per-view state, navigation history, caching and mouse handling. By the author's account, they archived the tool and started again, rewriting it by hand in Rust.

They're not alone. Earlier in the year, Mo Bitar published "After two years of vibecoding, I'm back to writing by hand" on Substack, describing how two years of agent-generated code had accumulated into what he called "pure, unadulterated slop" that he refused to ship to paying users.

Two posts don't make a movement. But these are two published accounts pointing at the same thing: code nobody fully understands.

> TL;DR

> - A May 2026 Hacker News post about abandoning a vibe-coded codebase cleared 1,000 points; Mo Bitar's Substack essay on the same theme came earlier in the year.

> - Stack Overflow's 2025 survey shows adoption still climbing even as trust falls. METR and DORA add the same tension from different angles.

> - The common thread is opacity. Governed no-code may be a different maintenance model, but only if a platform's controls are real.

How did we get here?

Vibe coding is the term Andrej Karpathy coined in February 2025 for building software by describing intent to an AI rather than writing code, and, in his words, "forget[ting] that the code even exists."

The adoption numbers are real. Stack Overflow's 2025 developer survey found 84% of respondents using or planning to use AI tools, up from 76% the year before, and 51% of professional developers using them daily. Y Combinator's CEO Garry Tan claimed on X that, for 25% of the Winter 2025 batch, 95% of lines of code were LLM-generated. That last one is a single self-reported batch figure, not a representative industry stat, but it shows how fast the practice normalised.

The trust numbers tell a different story. In the same survey, only 29% of developers said they trust AI output, down from 40% in previous years, and favourable sentiment fell from 72% to 60% year over year, per Stack Overflow's own write-up.

Is "almost right" actually worse than wrong?

The single biggest frustration in that survey, cited by 66% of developers, is "AI solutions that are almost right, but not quite." Almost-right code is its own kind of misery. Broken code gets fixed and you move on. Almost-right code compiles, passes the happy path, looks sensible, ships, then fails at 2am on an edge case nobody understood because nobody read what the AI had written.

A randomised controlled trial by METR, published in July 2025, had 16 experienced open-source developers work on real issues from their own repositories. When they used AI tools, they took 19% longer to complete tasks, with a confidence interval between +2% and +39%. They expected AI to speed them up by 24%, and even after the slowdown, still believed it had sped them up by 20%. In February 2026, METR published follow-up data on late-2025 tools, and was careful about what it showed: for returning original participants it estimated a speedup of -18% (confidence interval -38% to +9%), and for newly recruited developers -4% (confidence interval -15% to +9%). Both intervals cross zero, and METR itself attributed the wide ranges to selection effects, including developers who declined to work without AI, calling the later numbers weak evidence. The finding that stuck is the perception gap: developers who were measurably slowed still believed AI was making them faster.

Google's 2025 DORA report linked AI adoption with higher delivery throughput and a negative relationship with delivery stability. The report's own framing is the useful bit: AI is an amplifier. It doesn't fix a team; it makes a good team better and a struggling team's problems louder.

What about the security cost?

In March 2026, researchers at Georgia Tech's Vibe Security Radar found at least 35 new CVE entries directly associated with AI coding tools, and estimated the wider ecosystem figure at roughly five to ten times that number, about 400 to 700. Attribution here is hard by nature: tools that leave no commit signature, like Copilot's inline suggestions, are nearly invisible. A separate scan by Escape.tech of 1,400 vibe-coded apps reported 2,038 critical vulnerabilities, more than 400 leaked secrets, and 175 instances of exposed personal data including medical records and authentication credentials.

None of this is a blanket indictment of every AI tool. It's a reminder that code optimised for "does it compile?" isn't optimised for "is it safe?", and that a pipeline that ships faster can also ship the wrong thing faster.

What happens when the wall arrives?

In the k10s account, the wall arrived after the project had grown, once the codebase is too big for any one person to hold in their head. The k10s author put it plainly: roughly 30 weekends on the vibe-coded version, then archived and started over. Their own summary was that AI writes features, not architecture, and that the longer it drove without constraints, the worse the wreckage got. That's one developer's account of one project, not a verdict on the whole industry.

Red Hat Developer's Todd Wardzinski put the core dynamic well in February: "When you vibe code, your instructions become obsolete the moment code is generated. The code itself becomes the only source of truth for what the software does."

To my mind, that's the real through-line. When you can't explain why your system behaves the way it does, you can't maintain it. My shorthand is: when you can't maintain it, you don't own it.

What does this mean for no-code?

Here's where it gets interesting for builders who don't want to write code at all, by hand or by AI. (Disclosure: nocode.tech is published by Stacker, the platform I'm about to mention, so read the next bit with that in mind.)

What these developers are reacting to is opaque code: behaviour no single person can explain, produced by a process that was asked to make it work, not to make it maintainable. Governed no-code platforms may offer a different maintenance model. In theory, your app is defined in a configuration layer you can read and reason about, while the runtime is the platform's problem rather than generated source you inherit. Whether a given platform actually delivers that is something to evaluate, not assume: it depends on the controls it really ships, its permission model, its audit trail and its versioning. Not every no-code tool has them, and some AI app builders generate code under the hood that's every bit as opaque as a 1,690-line struct. Stacker, which publishes this site, describes itself as built around that configuration-led idea, and the same caveat applies to it.

My read: a useful test is whether you can point at your app and say exactly what it does and why, whether you got there writing clean code by hand or configuring a platform whose controls you've actually evaluated.

Is this a permanent shift or a wobble?

The enthusiasm cycle around AI coding has run wonder, adoption, disillusionment, recalibration. In my read, these accounts point to a more cautious phase now.

These two accounts do not describe a total rejection of AI. The k10s author's recommendations focus on writing down architecture and keeping view/state boundaries explicit.

The takeaway

For the rest of us, the builders who picked no-code before AI coding was an option, the takeaway is simpler than the doom-cycle headlines suggest. This isn't a rejection of building without code. It's a rejection of building without understanding. Whichever tool you choose, hold on to the part where you can explain what your software does and why.

Want to read
more articles
like these?

Become a NoCode Member and get access to our community, discounts and - of course - our latest articles delivered straight to your inbox twice a month!

Join 10,000+ NoCoders already reading!