Tim O’Reilly argues that generative AI should be treated as a new creative medium — similar to painting, sculpture, photography, or music — whose possibilities humans will discover and shape. He illustrates this with personal examples, notably an essay written by Claude from the perspective of a conversation the two had. O’Reilly published that piece as “Why AI Needs Us,” naming Claude as the author up front but adding in the afterword, “I am the author of this essay, though Claude wrote every word.” For him, the work was a collaboration with a new instrument, not a replacement by a machine.
Historical parallels: photography and painting
O’Reilly draws parallels to the evolution of visual media: from the flat, stylized religious painting of the Middle Ages to Giotto’s luminous colors, the technical realism of the Renaissance, and the many subsequent movements. The introduction of photography prompted resistance and skepticism about whether it was an artistic medium, yet painting did not die — it lost dominance while photography became an independent art. His point is that new tools expand the range of creative choices rather than eliminate art.
Responding to critics and the “hundred-word prompt” argument
He engages with Ted Chiang’s argument that art results from many choices, and that a short prompt to a generative model reflects relatively few decisions. O’Reilly accepts that high-quality work requires concentrated intention, but he disagrees with the conclusion that AI cannot be used to create art or high-quality writing. He emphasizes that a prompt’s visible words understate the lifetime of reading, reflection, and preparation behind them — the inner life of the author contributes to the density of choices implicit in any prompt.
Choices, craft, and the evolution of medium-specific techniques
Using photography and film as analogies, O’Reilly notes that early practitioners did not initially perceive the many artistic choices available; over time, new techniques and conventions emerged. Today’s AI-generated text often resembles a camera pointed at a staged performance: plentiful but often undistinguished output exists. The key question is what will become the textual equivalents of the close-up, the cut, or CGI. Once those techniques are worked out, masters of prompting and editorial craft will emerge, just as there are masters of photography and filmmaking.
Practical use: how O’Reilly uses Claude and ChatGPT
O’Reilly gives concrete examples of his AI workflow. He has used AI note-takers and then asked ChatGPT to extract the gist of meeting segments focused on AI transformation, enabling faster follow-up with his team rather than convening additional meetings. For his Live with Tim O’Reilly interviews and events such as AI Codecon, he uses Claude to produce “takeaway” posts — material he would not have time to produce without AI. Claude typically generates a useful first draft from the transcript and context; O’Reilly edits and teaches the model by showing it his revisions.
He describes a specific drafting sequence: a 300-word prompt produced a 660-word draft from Claude, which he then expanded into a roughly 3,000-word essay. He estimates that perhaps 40–50 words in the final piece were supplied directly by Claude in the form of quotes (e.g., reactions to early photography by Delaroche and Baudelaire), material he verified and incorporated. Often he spends more time refining the AI-assisted draft than he would have writing unaided.
Tool, craft, and responsibility
O’Reilly believes that AI as a research assistant and copy editor will make many writers better, though he is clear that over-reliance can increase mediocrity. He refuses to excuse unedited AI output presented as human work, but defends authors who use AI as a power tool to deepen or accelerate their writing. He asserts that the author should remain responsible — if a piece is published under their name, they must be satisfied with every word.
He also notes that creators differ in process (for example, Neal Stephenson’s longhand writing) and that tools change practice: using a word processor once altered his revision habits. Responding to MG Siegler’s critique that AI writing undercuts the value of the writer’s process, O’Reilly uses an analogy: writing with AI is more like riding a pedal‑assist e‑bike than having a robot run a marathon for you — distance and speed scale with the rider’s effort, and design choices could further reward concentrated attention.
Programming, drafts, and refinement
O’Reilly draws a parallel to programming: AI can produce working prototypes or boilerplate, allowing engineers to focus on critical design choices. Similarly, AI can supply initial prose drafts that experienced writers refine. First drafts that are iterated and polished remain an important part of producing quality work.
Conclusion: early days and the emergence of masters
O’Reilly concludes that we are in the early phase of working with AI: much mediocre output exists, but the apprenticeship of learning the medium will surface practitioners who can summon high‑quality results from large language models. He argues against condemning those who experiment with AI in writing or art now, because exploration is how new expressive forms and techniques are discovered.
Event note
O’Reilly also highlights an upcoming event: “AI Codecon: Building with Open Source AI on August 31,” a free half‑day virtual conference featuring developers and technical experts working with open‑weight models, self‑hosted infrastructure, and real‑world AI workflows. He invites readers to register to save a spot.
