On Making: Beej's Reflections on AI‑Generated Work and the Sense of Authorship

Beej feels uncomfortable claiming authorship of AI‑generated work

Beej states that he does not feel like he made anything when he has Claude or other LLMs produce code, art, or a deck for him, and that he prefers to say “I had this built for me” rather than “I made it”. He explains that completing projects is enjoyable, but having others complete projects he initiated is far less fulfilling to him.

The core distinction is between making and asking to be made

Beej concludes that the difference lies in whether he is personally doing the work or merely requesting someone else to do it. He writes that he feels pride when he writes C code, swings a hammer, or runs an assembler, because those activities feel like making. When he asks an LLM to write software, build a deck, or paint a picture, he feels he is asking others to make it, which carries little weight in his sense of making.

Hand‑written flash‑card system illustrates his pride in personal effort

Beej describes building a simple flash‑card system for his wife by hand, using only a little help from Claude to learn how to fetch CSV data from Google Sheets. The final program consists of 177 lines across JavaScript, CSS, and HTML. He notes that he can put his name on that code and say he made it, and that he is infinitely more proud of this modest, hand‑written code than anything he has had Claude write, because he cannot feel proud of the latter.

Prompting requires skill but is still a form of direction, not making

Beej acknowledges that effective prompting involves vision, judgment, communication, and prompting skill, and quotes Eisenhower: “Leadership is the art of getting someone else to do something you want done because he wants to do it.” However, he maintains that prompting is akin to asking someone else to make something, and therefore does not satisfy his personal sense of making.

Using a compiler still feels like making, while AI feels different

Beej observes that when he writes C code and lets a compiler produce assembly, he feels he wrote the program, even though he did not write the machine code. He argues that the translation from source to assembly is a strict, mathematically precise process comparable to using a hammer to drive nails, and therefore does not diminish the act of making. In contrast, he finds the relationship between prompts and AI‑generated output less predictable and less transparent, which makes it feel more like asking than making.

Community reactions highlight a split in perspectives

Commenters offer varied views on whether AI‑assisted work can be considered making.

• One commenter (@planb) says they can take pride in something made with an LLM even without writing code, because the end product would not exist without their initiative, comparing it to hiring a landscaping firm for a garden.

• Another (@layer8) argues that the key difference is the ability to reason about how input changes affect output: with a compiler they can predict misbehavior from source code, but with vibecoded programs they cannot reliably trace behavior back to prompts.

• A third (@unshavedyak) describes LLMs as a software equivalent to a 3D printer, noting they enable projects they lack time for, but they do not take pride in vibe‑coded output, only enjoyment of using the result.

• @bufbupa frames the issue in terms of transferable understanding: if one could reproduce the result without the AI (perhaps more slowly), they have made it; if the AI managed complexity and they internalized nothing, the work was commissioned, not made.

• @jvanderbot asks whether a machinist using a CNC machine still counts as a machinist, suggesting that reliance on automated tools does not necessarily erase the identity of the maker.

• @haickernews counters that a human in the loop guiding the process is 100% making, citing directors, collage artists, and fishermen as examples where the initiator is considered the maker despite delegating execution.

• @guybedo describes feeling like a team manager when using LLMs, enjoying the ability to build ideas that would otherwise require a large team, even if the code is less pretty and the process less fun.

• @e808 reflects on the value of AI‑accelerated projects, questioning what constitutes the value—the original idea, the prompt, token usage, or hours spent—and notes they enjoy being able to bring long‑held ideas to life quickly.

Beej leaves the question open, suggesting the line between making and asking is unclear

Beej concludes by noting that the boundary between making and asking to be made is not clear, quoting Justice Potter Stewart: “I know it when I see it.” He presents examples of what he considers making (writing C, swinging a hammer, using an assembler) and what he considers asking (requesting software, a deck, or a painting from another party). He acknowledges that examining the gray area makes him more confident that prompting AI can be an act of creation, yet he still feels unable to say he hammered a nail if an AI‑powered hammer does the striking.

He closes with an excerpt from The Hitchhiker’s Guide to the Galaxy about Deep Thought’s long computation, hinting that the act of setting a process in motion—whether by hand or via AI—may itself be a form of making, but he does not resolve the tension.