Il codice è solo un dettaglio? — Salvatore Sanfilippo
Is Code Only a Detail in Software History?
Why agent automation shifts attention from implementation effort toward software ideas, design, and architecture
In this Italian-language response to the claim that saying code was never the hard part insults programmers, Salvatore Sanfilippo separates implementation difficulty from conceptual invention. Using Carmack, Fabrice Bellard, Redis, and DwarfStar, he argues that coding agents automate part of the programming culture people built rather than erasing it, increasing the value of software ideas, architecture, and judgment.
Salvatore Sanfilippo responds to a widely shared essay about whether saying code was never the hard part devalues programmers. He rejects both the claim that implementation is trivial and the idea that implementation effort alone explains software's value.
His distinction is between implementation effort and the conceptual choices that decide what a system should be. Coding agents can automate part of the former because they inherit a programming culture built by people; that makes product intent, design, architecture, experiments, and judgment more consequential, not less.
Key Points Covered
- Code is a means inside the larger software problem: Sanfilippo says the disputed phrase need not mean programming was easy. In his framing, software also includes deciding what to build and designing how it should work; once a feasible goal and sufficient resources exist, implementation is difficult work in service of that goal 00:57–03:03(opens in a new tab).
- Implementation skill and conceptual direction reinforce each other: John Carmack's programming ability mattered, but Sanfilippo argues that the same ability without a vision for the games would not have produced id Software's work 04:08–05:06(opens in a new tab).
- Fabrice Bellard exemplifies exceptional implementation leverage: QEMU and FFmpeg required unusual technical skill and made important capabilities broadly available; QuickJS is his further example of a compact implementation. Sanfilippo nevertheless distinguishes these contributions from originating concepts such as virtualization or media compression, which predated those implementations 05:06–08:10(opens in a new tab).
- Redis is his counterexample from personal experience: He presents the database's in-memory model, fork-based persistence, replication strategy, and data-structure interface as the decisive ideas. The code was substantial work, but he locates the project's pivotal moments in those design choices 08:10–10:06(opens in a new tab).
- Automation depends on programming culture rather than nullifying it: Because stable code was difficult to produce, people built a large programming culture around it. Sanfilippo argues that language models can automate some coding precisely because people created that culture 10:06–11:09(opens in a new tab).
- Cheaper implementation raises the relative value of invention: He divides programming into implementation and conceptual work, then argues that reducing manual implementation effort puts more emphasis on program design, architecture, and genuinely new capabilities 11:09–13:16(opens in a new tab).
- Coding agents change who can realize ambitious systems: Sanfilippo contrasts programmers who could implement assigned work but lacked a larger product idea with people whose ideas exceeded their implementation capacity. He uses his roughly 150,000-line DwarfStar project as an example of automation making a previously impractical build possible 12:12–14:20(opens in a new tab).