Hank Green took some heat recently for using artificial intelligence in his work. Vanity Fair framed the controversy around the anger of some fans who believed his use of AI crossed a line between technological assistance and creative authenticity. One Hank would like to stand up for another.
The criticism raises a larger question that writers, researchers, artists, programmers, teachers, and nearly every other knowledge worker will soon find impossible to avoid. Will your customers pay the additional cost required for you not to use the most modern tools available? That is really what we are talking about.
Consumers have always made choices between handmade products and products made with varying degrees of automation. A handmade dining table can command a premium over one produced in a factory. A hand-thrown ceramic mug can cost substantially more than a mass-produced alternative. Buyers sometimes value the process enough to pay for it, and nothing is wrong with that.
The same principle becomes more complicated when applied to intellectual and creative work. Most people do not object when a writer uses a computer instead of a typewriter. Few readers demand that a researcher search a physical card catalog instead of Google Scholar. Photographers routinely rely on autofocus, image stabilization, exposure algorithms, and digital editing software. Musicians use digital workstations that automate tasks once requiring expensive studios and specialized technicians. Each technology changes the amount and type of human labor required to produce the final product, and artificial intelligence is another step along that continuum.
Green’s case is particularly interesting because the underlying concern appears to be less about technology than authenticity. The Vanity Fair article describes a dispute over Green’s use of ChatGPT in research and notes that Green acknowledged relying too heavily on AI-generated notes. He nevertheless disputed the suggestion that a controversial phrase itself had been written by ChatGPT (Carlstrom, 2026).
That distinction matters. Using artificial intelligence to search, summarize, organize, challenge, prototype, edit, or generate possibilities is not the same thing as asking a machine to create a finished work and putting your name on it. Those activities often get collapsed into the single phrase “using AI,” and that phrase is rapidly becoming almost meaningless. Talking about “using AI” today is nearly as imprecise as saying someone “used a computer.” The important questions are what the tool did, what the human did, what was verified, what was changed, and who ultimately accepted responsibility for the finished product.
Criticism aimed at accuracy is much easier to defend than criticism aimed at method. A creator who publishes inaccurate information owns the mistake, regardless of whether the error originated with an AI system, a research assistant, a spreadsheet, or the creator himself. Accountability should remain human, and tool purity is something else entirely.
My own experience with this problem is recent, and it happened here. Indianapolis is not the city most people picture when they imagine knowledge work, which says more about the picture than the place. Green’s own company employs writers, producers, and researchers a few miles from where I am typing, and they are hardly alone. Last week I published Bioinformatics with AI: A Modern Introduction, a fourteen-chapter textbook that took six years to write, and artificial intelligence was part of the process. I am quite comfortable saying that I probably would never have finished the book without it. The AI did not spend six years acquiring the knowledge behind the book. It did not decide that the book should exist. It did not assume responsibility for what appeared on the page. It helped make completion possible.
Then I gave the textbook away, and that little economic detail matters. A creator can spend hundreds or thousands of hours producing something useful without having a realistic path to recovering the full cost of that labor. Educational materials, independent journalism, open-source software, public-interest research, small creative projects, and countless other forms of intellectual work routinely exist under exactly those conditions. Automation can make those projects cheaper, and sometimes it does something even more important.
Automation can make the difference between a project existing and never being made at all.
Green reportedly explained that part of his attraction to AI came from wanting to make more things. He described the ambition plainly, saying he wants “to make too much stuff” and believes it is possible (Carlstrom, 2026). That seems remarkably ordinary to me. Creative people generally want to create. Researchers want to research, programmers want to build, teachers want to teach, and small businesses want to serve more customers without multiplying their expenses at the same rate. Modern tools make that possible.
Competition makes the issue harder to ignore. One writer may choose to perform every research task manually. Another may use AI to locate sources, summarize documents, compare arguments, transcribe interviews, and identify holes in a draft. The first writer is certainly free to work that way, but the market is not obligated to compensate that writer for choosing the more expensive production method. That is the uncomfortable economic reality beneath much of the current AI debate.
Customers generally want better products, lower prices, faster delivery, or some combination of the three. Businesses and creators that discover tools capable of improving those dimensions gain an advantage. Competitors can refuse those tools, but they must eventually explain who will pay for that refusal. Some customers will.
There will always be a market for handmade furniture, analog recordings, hand-bound books, artisan food, and other products where the method of production is itself part of what people value. Creative work may develop similar niches. An author might someday advertise a novel as entirely human-written, just as a furniture maker advertises a table as handmade. That would be perfectly legitimate. The mistake would be assuming every creator must operate that way.
Nothing is wrong with demanding handmade work. Something is strange about demanding handmade work while also expecting automated prices, automated speed, and automated volume.
Hank Green should be responsible for what Hank Green publishes. So should I, and so should everyone else. That standard is demanding enough. Creators should verify facts, disclose AI involvement when it materially changes the nature of the work, protect the trust of their audiences, and remain accountable for what they put into the world. They should not be required to pretend that the tools of 2026 do not exist.
The competitive question is no longer whether professionals will use artificial intelligence. The important question is whether they can use it without surrendering judgment, accountability, originality, and the qualities that made their work worth consuming in the first place.
Reference
Carlstrom, M. (2026, August 14). The brief wondrous cancellation of Hank Green. Vanity Fair. https://www.vanityfair.com/story/hank-green-ai-youtube
Appendix: The AI Tools I Currently Use
The phrase “using AI” can hide an enormous range of activities, and my own working environment makes that clear. What follows is an inventory taken in August 2026. It will be out of date quickly, which is itself part of the point.
Local and self-hosted models. These run on my own hardware and handle work I do not want leaving the building, along with anything high-volume enough that per-token pricing would be punitive. The general-purpose models include muse-glimmer 30b, ornith 35b, gemma4 at 12b and 31b, gemma3 27b, qwen3 8b, qwen3.5 35b, qwen3.6 at 27b and 35b, llama3 and llama3.1 at 8b, phi4 and phi4-mini, gpt-oss 20b, nemotron-3-nano 30b, laguna-xs 2.1, and glm-4.7-flash. Coding work runs through qwen3.5-coder 35b, qwen2.5-coder 32b, and devstral-small-2 24b. Reasoning-heavy tasks go to deepseek-r1 32b and cogito 14b. Retrieval and search indexing use qwen3-embedding 8b. Several of these are held in multiple quantizations depending on whether a given job needs speed or fidelity.
Cloud and API models. These handle work where the local models are not strong enough or where I need a second opinion from a different architecture. The current set includes DeepSeek V4 Flash, Kimi K3 and K2.5, Gemini 3 Flash Preview, Claude Opus 5, GPT-5.4 Mini, MiMo-V2.5 and V2.5-Pro, MiniMax M2.5, and GLM-5.
Platforms and infrastructure. ChatGPT, Claude, Gemini, and Perplexity for interactive work and research. Midjourney for image generation. Ollama and MLX for local serving. OpenRouter for routing across cloud providers without maintaining separate accounts.
That list is not included to suggest that more AI is automatically better. It is included because it shows how misleading the binary question has become. I do not use AI in one singular way. I use different systems for different tasks, under different cost structures, on different hardware, with different privacy implications, strengths, and failure modes. Choosing among them is itself the work, and it is increasingly what modern professional practice looks like