OpenAI Academy published a case study on October 1, 2026 about journalist and technologist Jaemark Tordecilla. He used Codex to build Working Draft, a tool that helps reporters cover a Philippine trial involving marathon sessions, many speakers, and testimony in multiple languages. The interesting part is not simply using AI to write a summary. It is connecting transcripts, timestamps, evidence review, and editorial judgment into a repeatable research workflow.
Working Draft identifies speakers in each day's transcript and adds timestamps that take a reporter back to the corresponding moment in the Senate video. It also summarizes the full session, lists witnesses and documents discussed, and produces an audio summary. Reporters can move from the summary back to the original video rather than treating generated text as the final record.
OpenAI Academy says Tordecilla built a usable site for the first day's proceedings in roughly a day and then spent about a week catching up on a month-long backlog. He can now process a new session in about three hours, with the last hour spent checking the material himself. That split is instructive: the agent reduces the cost of building and organizing the tool, but verification remains a required part of the workflow.
The case study also says he used AI coding tools to turn the Philippines' 2025 national health survey into a dashboard with maps, charts, and plain-language summaries checked against the source data. In another project with Rappler, Codex and ChatGPT are being used to search roughly 18,000 local-government audit reports for potential story leads. These are forms of data organization and exploration, not a transfer of editorial responsibility to the model.
The name Working Draft captures the product's intended status: material to check before publication. When reporting involves testimony, documents, and public figures, a wrong speaker label, timestamp, or over-simplified summary can change a reader's understanding. Making every summary traceable to source video and keeping final verification with a person is more important than maximizing generation speed.
The workflow is practical. Codex builds and updates the tool, models organize transcripts and search records, and the journalist decides what deserves attention, checks evidence, resolves ambiguity, and chooses what to publish. The value is not full automation of journalism. It is allowing a domain expert to create a specialized research interface with far fewer engineering resources than a conventional newsroom project would require.
The portable lesson for other content teams is to retain source locations, item-level checks, replayable processing, and a clear draft state. AI can make a large archive searchable and turn long recordings into leads, but public work still needs a separation between original evidence, human judgment, and publication responsibility. Working Draft is better understood as a verifiability-first agent workflow than as an AI-replaces-journalists story.
