Media's and Influencers' Narrative about AI Vibe Coding is Nonsense | Carl Brown (Internet of Bugs)

Summary

Carl Brown argues that media and influencer coverage has overstated what AI-assisted “vibe coding” can accomplish. AI can speed up an early prototype—often building the simplest 80% of a project quickly—but debugging, adding complex features, and finishing the final stretch can take much longer or fail altogether. When developers do not understand AI-generated code, fixing problems can be especially difficult: the code may use unexpected approaches, and changes can introduce new bugs.

Adobe-Style Alternatives and Product Assessment

The discussion challenges headlines suggesting that a small team used AI to recreate or decompile Adobe products and produce ready-to-use replacements. Brown says the applications he understands to be involved are pre-alpha or very early prototypes, not substitutes for Adobe’s mature tools.

• Brown tried FilmCraft, a Premiere-like editor, to work on a podcast episode. He could not use familiar editing workflows and features, including options he expected to find when working with clips.
• The host tried the Photoshop-like alternative and described it as nonfunctional or very primitive.
• Both speakers consider the early progress impressive for a short development period, but say it falls far short of replacing professional software.
• The host also discusses Affinity, which he says became free after Canva acquired it. He finds it usable and stable, though not as capable as Photoshop or Illustrator in his experience.

The speakers say Adobe’s appeal is not only its feature set. Users value reliability, established workflows and shortcuts, and the fact that Adobe skills are commonly requested by employers. Businesses may also pay vendors so they can focus on their own core work rather than maintaining software themselves. Existing open-source alternatives show that software can be reproduced without automatically winning users or market share.

AI Coding, Debugging, and Software Maintenance

Brown describes AI as a tool that amplifies developers rather than replacing software engineering. It can make straightforward work easier, but it does not remove the need for people who can understand architecture, diagnose bugs, and maintain a codebase. Vibe-coded projects may reach a quick first draft, then encounter “whack-a-mole” debugging: fixing one problem creates another, making the software difficult to finish and ship.

He compares inheriting unfamiliar AI-written code to taking over a codebase written by someone else: names, structure, and design choices may not reflect how a human maintainer expects the software to work. His broader point is that getting 80% of a project done is not the same as being 80% finished.

Reverse Engineering Versus Decompiling

Brown distinguishes two concepts often conflated in coverage:

• Decompiling attempts to translate a compiled program’s low-level machine instructions into a more human-readable form. He says such tools have existed for a long time; AI can make them easier for less-specialized users to operate.
• Clean-room reverse engineering means building software from observable behavior, specifications, interfaces, and tutorials without examining the original source code or executable.

Brown’s understanding is that the Adobe-style projects were clean-room implementations, not decompilations. He notes that decompiling commercial software can raise serious intellectual-property issues. He also discusses decompiling in security research and piracy, including claims about PS5 jailbreaking and running older games or PS5 titles on PCs; he presents some of those details as reports he has seen rather than as something he personally verified.

SaaS, Competition, and Business Models

The speakers discuss why companies often license or subscribe to software instead of building and maintaining their own. An alternative may be technically possible but still require staff time, training, support, and migration away from established workflows. Brown says AI may lower the cost of creating competitors, but it does not erase those business obstacles.

They also consider whether the shift toward subscription or cloud-only software could create openings for local applications. Brown cautions that demanding tasks such as video editing involve large files and frequent interactions; network latency can make remote editing frustrating even on a fast local network. If cloud software performs poorly, locally run competitors could benefit, though dominant companies may use pricing power or acquisitions to limit competition.

Other Technology and Media Analysis

Brown argues that AI has an asymmetric role in cybersecurity: an attacker can try repeatedly and needs only one successful attempt, while defenders must catch every significant threat. He recommends established security tools and practices, with AI used—if at all—as a supplement rather than the sole defense.

The discussion also examines AI’s impact on hiring. AI-generated résumés and cover letters, AI screening systems, and automated interviews can weaken the signals employers and applicants once used to assess one another. The speakers criticize media and influencer narratives that present AI as either a universal replacement for workers or an existential threat, arguing that its usefulness depends on the task and context.

Reviews, Guides, and Tutorials

• Product tests discussed: Brown’s FilmCraft editing test; the host’s test of the Photoshop alternative; the host’s assessment of Affinity.
• Conceptual explanation: Brown gives an overview of how compiling and decompiling work and distinguishes decompiling from clean-room reverse engineering.
• No step-by-step software tutorial is presented.

Main Speakers and Sources

• Carl Brown — veteran software engineer, author, podcaster, and creator of Internet of Bugs; provides the main technical analysis.
• The Retort host — interviews Brown and shares personal experiences with Adobe, Affinity, and AI-related media coverage.
• Other named references include Jeffrey Funk, mentioned in relation to earlier AI discussions, and examples of products and organizations such as Adobe, Canva, DaVinci Resolve, Figma, Sony, and OpenAI.