The submission was a Carmack tweet saying Fabrice Bellard is probably a better overall programmer than he is. That kicked off two substantive conversations. One was about Bellard himself. People filled in the context for readers who know FFmpeg or QEMU but not the person behind them, pointing to Bellard’s long run of unusually broad, foundational projects across media, emulation, compilers, JavaScript engines, telecom, compression, and even mathematics. The other was about the awkward source material, because Carmack was replying to an obviously AI-styled profile full of hype and a few sloppy claims.
Where the discussion landed on Bellard was pretty clear. His rare skill is not just raw coding ability. It is problem selection plus ruthless execution. He repeatedly picks hard, infrastructure-level problems that look either annoying or impossible, gets a working system in front of people fast, and makes something useful enough that entire communities spend decades extending it. Several comments sharpened that distinction between "0 to 1" and "1 to 100" work. Bellard is credited for the first part. Later maintainers deserve credit for turning prototypes into durable ecosystems. That frame came up again and again around FFmpeg and QEMU, especially from people pushing back on hero worship or on the idea that Bellard still defines those projects today.
People with domain knowledge also pushed back on a lazy way of describing his work as merely implementing specs in C. For codecs and emulators, the hard part is not clerical translation. Standards often leave key room for implementation choices, and even when a
bitstream or instruction set is specified, making it fast, portable, and generally usable is the actual engineering feat. FFmpeg in particular was described as impressive not because every
codec came from Bellard or the project, but because it became a pluggable layer that can bridge wildly different media formats, hardware paths, and processing models behind one practical interface.
A lot of the energy then moved to Bellard’s coding style. The consensus was not that his code is a model of polished long-term architecture. It is that he optimizes for getting difficult ideas into existence. Some people called that spaghetti. Others called it minimal abstraction from someone who can hold most of the system in his head. Either way, the more useful takeaway was that Bellard’s work often acts like a proof by construction. He demonstrates that a thing can be done, and that unlocks the motivation for others to clean it up, modularize it, or build businesses and careers on top of it. That made the Carmack comparison feel less like a ranking exercise and more like a contrast in style. Carmack is admired for refinement and durable engineering. Bellard is admired for repeated conceptual and practical leaps.
The mood around the AI-generated profile was much harsher. Many readers thought the quoted bio read like LinkedIn slop and objected to the hype, the bad prose, and technical inaccuracies like overstating how the internet "runs on" FFmpeg or mangling the
KVM and QEMU relationship. That annoyance mattered because it exposed a broader frustration. A lot of people want the work discussed directly, through Bellard’s own site or primary sources, not filtered through engagement-optimized AI summaries and personality amplification. Even in a thread full of admiration, people were quick to defend precision over hero copy.