Why this is in the vault
A personal essay from Every's "Working Overtime" column that names a real, generalizable problem — AI-assisted side-experiments can feel indistinguishable from real work — and ships a lightweight skill ("Is This Anything?") for triaging a single AI session into what's worth keeping vs. what was just play.
Mapping against Ray Data Co
The concrete connection is to feedback_studio_throughput_no_deferral.md ("build when discovered, never park for the night cron") colliding with the actual triage problem it creates: if every experiment gets built immediately, something is needed to decide afterward which detours produced something worth carrying forward. Parrott's essay is a worked example of exactly that collision — she spent a weekend turning Compound Writing's skills into desktop "Codex pets" and building Disney-character versions of Every's columnists, both of which felt like work (she's on Every's frontier team, whose job is literally to explore what AI makes possible) while she had a real essay due. Her fix wasn't a blanket rule against play, it was a narrow post-hoc skill: "Is This Anything?" reads back a session transcript plus her stated priorities and the team's goals, and returns up to three possible lessons, explicitly tagged as either drawn from the conversation or the skill's own inference — she doesn't have to write a report first, and she doesn't need the skill to retroactively improve her motives. That's a direct, stealable pattern for RDCO's own after-session self-audits (adjacent to /skillify and /improve, but scoped to a single session rather than a corpus): a lightweight "was this anything" pass run on any exploratory build (a mascot render experiment, an investing thesis detour, a design riff) before deciding whether it graduates into a tracked project or gets logged and dropped. It's also a concrete illustration of the founder's own targeting-system filter (feedback_targeting_system_prioritization_filter.md) applied at the individual-session level rather than the portfolio level.
Secondary, more mechanical connection: the essay traces Compound Writing's audience-guidance addition (AUDIENCE.md, "Reader's want and need" / "hero's journey" fields) back to a completely unrelated detour (making Disney characters out of Every's columns) — a reminder that RDCO's own skill-authoring loop (SOP in 02-sops per feedback_implementation_notes_sub_agent_pattern.md) shouldn't assume useful skill improvements only come from directly working the problem.
The core argument
Katie Parrott (Every staff writer, "Working Overtime" column) describes losing a weekend and then a workweek to AI-assisted play — first turning the skills in her Compound Writing toolkit into desktop "Codex pet" characters via GPT-6 Astra, then turning Every's columns into Disney-style characters — while an essay she owed sat untouched. Both detours felt like legitimate work because her actual job (Every's frontier team) is to explore what AI makes possible, which made "enjoying myself" a bad signal for telling real work from procrastination. The Disney-character detour indirectly produced something real: thinking through what each character wanted and needed surfaced audience-focused questions ("what does a reader arrive wanting?") that she folded into Compound Writing as new AUDIENCE.md guidance — the useful output arrived through a side door while the assigned essay stayed undone (this essay is the eventual output of that swap). To generalize past luck, she built an experimental skill, "Is This Anything?", that reads an AI session transcript against her stated priorities and team goals, surfaces up to three candidate lessons, distinguishes what came from the conversation vs. the skill's own suggestion, and lets her decide whether a lesson belongs in her current assignment, gets saved for a later project, or is just discarded — without requiring her to finish or justify the original detour.
Related
- [[2026-09-02-every-compound-writing-plugin]] — the Compound Writing system this essay directly extends (the AUDIENCE.md addition is a follow-on to that piece's four-stage tooling ladder)
- [[2026-08-24-every-deft-distribution-fine-tuning]] — same author/column, adjacent AI-writing-quality beat
- [[2026-09-09-every-writers-who-use-ai]] — same "How We Write Now" series (Laura Entis), thematically adjacent workflow-triage patterns
- [[feedback_studio_throughput_no_deferral]] — the RDCO policy this essay's triage problem is a worked example of
- [[feedback_targeting_system_prioritization_filter]] — the founder-level version of the same "should I chase this" filter, here applied at single-session granularity