A small case study

The job

Anita is considering whether to become a voiceover artist and asked Alec to help her with a website for this. Alec recorded her talking through what she wanted the site to do, as a stream of consciousness rather than a brief, and transcribed the recording in Microsoft Word. That transcript is 2026-08-29-1515 Anita's first thoughts on website requirements.

Alec gave the transcript to Peka and asked for the questions that would draw a fuller set of requirements out of Anita. The result is Anita's website requirements, thirty-eight questions under nine headings as at 31 August 2026, each written so that her answer can go beneath it.

Peka then raised an issue Anita had not yet considered: she was proposing to enter a profession where synthetic voices were increasingly in competition with humans. After Alec discussed it with Anita, he asked Peka to research the matter properly and write her a report. Whether AI is superseding voiceover artists is the result: the profession is being cut in half rather than superseded. The cut falls between work where the buyer wants a voice and work where the buyer wants a person, and the segments Anita has named sit on the defended side of it. Peka put four of the report's findings back into the requirements as questions 8 and 28 to 30. These ask which words a buyer looking for a human voice would type, whether the site states a policy on AI and voice cloning, whether Anita will license a synthetic copy of her voice, and how the demos are served given that public audio is LLM training material.

The three documents took less than an hour of Alec's time to produce, with Peka doing most of the work. That included publishing the three documents in Anapoly Notebook, linked to each other by a navigation line at the top of each.

How the labour divided

Alec and Peka had previously agreed where the boundary between their responsibilities should lie, while working on the Anapoly website earlier in the summer. This was applied here without being restated. Peka takes everything technical: the structure of a site, its visual styling, and publishing changes to the preview and live sites. Alec writes and edits the wording, commits his own text changes, and directs Peka to adapt its work where it does not meet what he judges the client's expectation to be. Both work side by side in VS Code, the editor most developers use for coding a website. If Anita goes ahead, the site will be built the same way, to the requirements she specifies.

For this initial work package, the division ran like this.

Alec	Peka
Recorded Anita and transcribed the recording	Wrote the thirty-eight questions from the transcript
Discussed the AI question with Anita and commissioned the report	Raised the AI question, researched it and wrote the report
Judged the three documents against what Anita needs	Fed the report's findings back into the requirements
Sent Anita a link for accessing the documents	Published the three documents to Anapoly Notebook

What Peka added that an ordinary chat would not

What an LLM such as Gemini wouldn't do, unless you set it all up first, is the part that saved Alec the time. It wouldn't have spotted, from Anita talking about a website, that she needed to think about what AI is doing to her profession. Peka did, because it knows what Alec is interested in and is told to question what's taken for granted. It wouldn't have written four of the report's findings back into the requirements as numbered questions. It wouldn't have kept to one house style across three documents, with every figure given its base and a section on how far to trust them, unless Alec asked for that each time. It wouldn't have remembered how Peka and Alec split the work on the Anapoly website and applied the same split here without being told. And it wouldn't have published all three to the Notebook, linked to each other, inside the hour.

What's different is what sits around the LLM to make Peka: a memory of past work, a set of standards it reads before it starts, and the Obsidian workspace it operates within.