<- all tokdocs

Spec-Driven Development vs Vibe Coding: Mario Zechner's Critique

By , founder of Mavric and Apso

View on TikTok ->

Spec-driven development has an AI agent work from a detailed written specification instead of loose prompts. In a TikTok from @wordman.dev, podcast host Jan-Niklas Wortmann and Pi creator Mario Zechner argue that a detailed spec only makes you an "enterprise vibe coder," and that having too much code to review is no excuse. Their worry about unreviewed AI code is backed by published data, but the clip's "specs or vibes" framing misses the review checkpoints that GitHub's Spec Kit builds in.

Where is the clip from?

The clip is a word-for-word excerpt from "Code Isn't Free - Mario Zechner on the Hard Truths of Coding With AI (creator of Pi)", episode 19 of The Weekly Dev's Brew, dated 12 June 2026. The TikTok was posted on 9 October 2026, four months later.

The caption names only Zechner, but the video shows the host, Jan-Niklas Wortmann, delivering the main lines. Wortmann's about page describes him as host of the show and a Solution Architect at Cursor. Zechner, on a webcam feed, agrees and adds the vibe coder comparison. Both men hold the same position.

What do they say about having too much code to review?

Wortmann says he keeps seeing the argument online: "well, we have too much code now, we cannot review it anymore." His response: "then how are you expecting to maintain it if you cannot read it? That doesn't make any sense what you're saying. That is just a little bit more sophisticated approach to vibe coding, but not much."

Zechner adds: "the vibe coder will just sit there with his voice dictation app and dictate what app they want at a very high level. And the enterprise vibe coder will first write a very detailed spec. But essentially it's the same fucking thing."

Who argues you should stop reviewing AI code?

The clip names nobody. The clearest public version of the position is Steve Yegge's. He told Tim O'Reilly in an O'Reilly Radar conversation on 12 March 2026: "If you're looking at your code, then you're in a Formula One race and you've parked your car and opened the hood...Code is a liquid. You spray it through hoses. You don't freaking look at it."

The Register's review of Gene Kim and Yegge's book Vibe Coding quotes Andrej Karpathy's framing: "Accept All. Always. Don't read the diffs anymore." The authors also report what went wrong, including an AI that "silently deleted or modified test cases, removing 80% of one large test suite."

Yegge's argument moves review from the diff to behaviour, with tests and outcomes doing the checking. The clip asks a fair question of that approach but does not engage with the answer.

Is spec-driven development the same as vibe coding?

Not as GitHub documents it. GitHub's Spec Kit announcement from 2 September 2025 calls specs "living, executable artifacts that evolve with the project." It has four phases:

  1. /specify writes a specification focused on user journeys.
  2. /plan writes a technical plan against your stack and constraints.
  3. /tasks breaks it into "actionable, reviewable work items."
  4. Implementation.

On the human's role it says: "Your role isn't just to steer. It's to verify." Vibe coding as Karpathy described it removes the verification step. Spec Kit adds one at every phase.

Zechner's stronger objection comes later in the full episode: "now you're not even writing the spec anymore you just vibe prompt your agent to write a very detailed spec and that's like a tower of turtles." That criticises how people use specs, not how the toolkit is written. Thoughtworks' Liu Shangqi makes a similar point, noting that "spec drift and hallucination are inherently difficult to avoid."

Is spec-driven development just waterfall?

The episode's chapter title is "Spec-Driven Dev Is Just Hyper-Waterfall." In it, Wortmann says "we literally introduced scrum" and corrects himself from "20 25 years ago" to "30 years ago." The correction is right. Jeff Sutherland and Ken Schwaber presented "The Scrum Development Process" at OOPSLA in 1995.

What does the data say about maintaining AI-generated code?

The clip cites no data, but published research mostly supports its worry.

  • GitClear reports, across 211 million changed lines from 2020 to 2024, copy/pasted lines rising from 8.3% in 2021 to 12.3% in 2024, while refactoring fell from 25% of changed lines to under 10%. A later report covering 623 million changes from 2023 to 2026 shows code block duplication up 81% and refactoring moves down 70%.
  • Google's 2025 DORA report surveyed nearly 5,000 technology professionals. 90% use AI at work and 30% report little or no trust in AI-generated code. AI adoption had a positive relationship with throughput and "a negative relationship with software delivery stability." DORA's explanation: "Without robust control systems, like strong automated testing, mature version control practices, and fast feedback loops, an increase in change volume leads to instability."
  • Stack Overflow's 2025 survey found 46% of developers distrust AI output accuracy, up from 31%. Its AI chapter says "The biggest single frustration, cited by 66% of developers, is dealing with 'AI solutions that are almost right, but not quite.'"
  • Faros AI, a vendor, reports from its own telemetry that under high AI adoption pull requests are 51% larger and 31% more merge with no review. Its data is not open to inspection.

Bottom line

The clip's core point holds: code nobody can read is code nobody can maintain, and the data shows AI-heavy teams losing stability. Its equation of specs with vibe coding ignores the review checkpoints in GitHub's Spec Kit, and DORA's findings point to testing and feedback loops as the deciding factor, not the methodology label.

Sources

Published October 9, 2026. Based on a TikTok by @wordman.dev. Full video notes (markdown)