Vibecoding Photoshop: Time and pressure
Rapid advances in large language models are enabling individuals to “vibecode” highly tailored clones of complex tools like Photoshop, raising questions about the future of commercial software, platform lock‑in, and professional workflows. Commenters debate whether most users will move from monolithic apps to personalized, task-specific tools, how sustainable this is given AI costs and quality pitfalls, and what it means for existing players like Adobe. The thread also touches on IP and ethics concerns, including AI-assisted code theft from projects like Photopea and the limits of protecting client-side software in an era of easy decompilation.
Personalized “Vibecoded” Software vs Monolithic Apps
- Many see AI-assisted “vibecoding” as enabling ultra-personal tools: a one-user “Photoshop” that has only the 10% of features they actually need.
- Some report already building niche tools (e.g., custom video players, Excel-like apps, internal Photoshop-like systems) in days or weeks.
- Others argue fragmentation is a problem: countless half-maintained clones instead of contributions to established open source projects.
Platforms, OSes, and Device Ecosystems
- If 90% of Photoshop/Word-like workflows can be replaced with personal apps, several posters question how much OS/platform lock‑in will matter.
- Counterpoint: hardware quality still matters (battery, trackpad, sleep/suspend), and some expect to keep using Mac laptops even if software “unbundles.”
Adobe, Photoshop, and Alternatives
- Strong hostility to Adobe’s pricing, bloat, and AI integrations; some hope vibecoded or open clones topple its dominance.
- Others defend Adobe’s “earned throne,” especially historically, and note professionals will keep paying for official tools and PSD compatibility.
- Browser-based Photoshop-like tools are cited as proof that one-person or small-team alternatives are possible even without LLMs.
Quality, Correctness, and Limits of AI Clones
- Experienced graphics developers warn that correct color spaces, resampling, and filtering are subtle and often wrong in existing code; LLMs reproduce these mistakes.
- View that casual users may not notice, but professionals will.
- Some say failures themselves will become future training data and quality will eventually “just work,” though others note much expert knowledge is closed and won’t enter training sets.
Economics, IP, and Cloud Shift
- Several expect binary “moats” to erode as LLMs help decompile and reimplement software; this may push more apps to cloud/streamed-only models.
- Skeptics point out that server-side rendering of heavy apps would be extremely costly.
- Debates over whether future AI access will remain cheap; some foresee rising token costs after current “subsidized” phase.
Broader Social & Creative Concerns
- Some imagine agent networks that generate custom videos, software, and even synthetic influencers on demand; others find this dystopian and socially isolating.
- One commenter calls AI art inherently disrespectful to human creativity; others lean into AI as tooling, not replacement.