The strongest reaction was that the browser-first design is the real breakthrough. People who have bounced off reverse engineering before said the usual blockers are ancient compilers, patched SDKs, and brittle local setup. Being able to open a tab and start diffing assembly makes the field feel accessible in a way most decomp projects do not. Several comments pushed the idea one step further: the natural next product is not only a course, but a web interface that lets people pick up unmatched functions from active projects and submit fixes without ever installing the toolchain.
The comments also clarified what this work actually looks like in practice. Matching decomp is usually done piecemeal, often one function at a time, with scaffolding that lets a growing mix of handwritten C and preserved assembly keep rebuilding the original binary. For that to work, you generally need the original compiler version and flags, and many projects aim for bit-for-bit equivalence except for clearly irrelevant sections like debug info. That is why the site is strict about a single wrong instruction counting as failure.
A second theme was where AI helps and where it stalls. People shared stories of frontier models doing strong first-pass reverse engineering, asset extraction, and platform-porting work on old software. But the post author's experience matched a lot of current reality: models can get functions to the 90 to 95 percent range, then fail on the last details that matter for a true match. That makes AI useful as an assistant for naming, probing runtime behavior, and getting rough structure, not a substitute for understanding compilers, calling conventions, and assembly quirks.
There was also a legal undercurrent. Some commenters warned that
decompilation projects sit in a gray area and that Nintendo in particular is aggressive. The more grounded reading was narrower: source-only projects that avoid shipping copyrighted assets or ready-to-run binaries have generally fared better than projects that distribute rebuilt binaries. Nobody treated this as settled law, but the practical advice was clear. Keep it educational, avoid monetization, and do not ship redistributable builds that invite a rights-holder response.
On the product itself, early users quickly found rough edges. One person could fake-match a lesson by exploiting the fact that only assembly equality is checked. Others hit a full compiler storage error, mobile layout confusion, and a lesson that felt too text-heavy too soon. Requests for an assembly primer, instruction reference, light theme,
ARM or x86 support, and a guide for starting a new decomp project from scratch all point to the same thing: the idea landed, and now the bottleneck is curriculum and UX polish rather than whether there is demand.