Inspiral Fieldwork / Case study · case study
How OpenAI Supports Inspiral Fieldwork
The concrete roles ChatGPT Work, connected context and Codex play in Inspiral fieldwork—and the human authority they do not replace.
OpenAI tools have helped ShiftBy do more of the research, synthesis, implementation and review work around Inspiral. They have not become the authority for what the work means, what is current or what may proceed.
The question
What has OpenAI materially contributed to Inspiral fieldwork—and where does that contribution stop?
The answer is not that AI built a system. It is more specific: connected working tools have helped inspect, retrieve, compare, draft, implement and challenge work that would otherwise be spread across documents, interfaces and code.
A working contribution ledger
| Tool or connected capability | Contribution in this fieldwork | Human authority retained |
|---|---|---|
| ChatGPT Work | Continuing research, comparison, synthesis and drafting across related fieldwork. | Evidence selection, claim boundary and consequential interpretation. |
| Notion app/plugin in ChatGPT | Retrieval of planning, evidence and specification context used to ground the work. | Whether retrieved material is current, applicable or authoritative. |
| Codex | Artefact inspection, dashboard and public-site implementation, testing and revision support. | Semantic meaning, implementation authority and release approval. |
This is a record of working roles, not a product stack. A connected tool does not make retrieved material true, current or ready to use.
An observed fieldwork correction
One observed correction shows this boundary in practice.
The tools helped inspect dashboard surfaces, retrieve related specifications and synthesize an initial public draft. A tool-assisted review then identified a material semantic error: the draft treated purpose-specific semantic profiles as though they belonged within ContextCore’s transformation. The accountable author accepted the correction.
ContextCore establishes a governed knowledge foundation. CanonLens later determines purpose-specific relevance and projects semantic profiles. The correction did not make the tool unsuccessful. It showed why tool contribution and authority must remain separate: a coherent draft can still cross a semantic boundary it is not entitled to cross.
The useful contribution is not automatic completion. It is increased capacity to make work inspectable, challengeable and revisable.
Trust remains only when someone can inspect the support, challenge the interpretation, correct the boundary and decide what may proceed.