The gap between what AI can demo and what it can deliver is growing, and this project is the clearest example we have seen in months. A developer used Claude Opus 5.5 to build open-source replacements for seven Adobe applications, then told two different communities two different stories about readiness. On Reddit, the projection was roughly 100% parity within a month. On Hacker News, the same developer admitted the project was not anywhere close to ready. The project's own roadmap puts it at 45% parity for real work, with an estimated 2,300 to 4,100 agent-hours remaining before it reaches full functionality. That is not a near-finished product. That is a prototype with a long road ahead.
We should be clear about what this means. The AI-generated code produced a compelling surface. It looks like the real thing in screenshots and quick demos. But the distance between 45% and 100% is not just more of the same work. The last 55% of a creative application, edge cases in file formats, color space handling, undo stacks, plugin compatibility, accessibility features, is disproportionately harder than the first 45%. This is the same lesson we see in other domains of tooling. Explore how dotnetup simplifies .NET management without admin rights shows how even a focused utility tool requires careful cross-platform engineering before it is ready for public preview. If a .NET version manager needs that level of polish, imagine what a Photoshop clone demands.
The developer's honesty on Hacker News is worth respecting. They corrected the record. But the original Reddit post already did its work: it generated attention, speculation, and investment of trust from people who heard "months away" and imagined "weeks away." The real takeaway here is about credibility, not capability. When you claim AI can build a full creative suite in a month, you are not just overpromising on a project. You are teaching users to distrust the next honest estimate. Compare this with how Four Phases to Master AI Product Management by 2026 frames the skill of evaluation: knowing what a model can actually do, not what it appears to do, is the core competency that separates useful AI adoption from hype-driven disappointment.
The specific number to watch is not the 45% parity figure. It is the agent-hour range: 2,300 to 4,100. That is a 78% spread. The developer cannot yet estimate the work within a factor of two. That uncertainty tells us more about the state of AI-generated software than any demo ever could. Until we see projects that can predict their own completion within a reasonable margin, the responsible position is to treat every "almost done" AI build as a starting point, not a finish line.