LO
CapabilityCognitive automation2026-09-10
OpenAI published a data-agent template whose own instructions require the metric definitions, trusted queries and corrections to be supplied by the team using it
Data analystoccupation page →
Event date / reported
2026-09-10
Evidence stage
CapabilityA demo, benchmark or paper shows the task can be done. Updates what the technology can do — not what employers will do.
Tasks this bears on
Writing queries and building dashboards
Translating 'how many users did X last month' into SQL, and wiring the result into a chart someone can refresh.
Automating✓ Evidence-backed
Knowing when the data is lying
Spotting the broken pipeline, the duplicated events, the timezone bug, the definition that changed in March.
Still human-led✓ Evidence-backed
Owning the definitions the tools rely on
Maintaining the metric definitions, the documentation and the semantic layer that make natural-language querying produce correct answers.
New task≈ Platform inference
Where this applies
A vendor walkthrough and reference implementation, not a report of use at any company — no employer is named as running it. Recorded as capability, which on this site never moves an assessment. What each task link rests on: the template's own instructions cover table discovery, read-only SQL and explaining the findings (writing queries); its prerequisites and step 3 require the team to supply metric definitions, trusted filters and previously reviewed queries, and step 9 saves a correction only when a person asks for it (owning the definitions the tools rely on); and the filters it prints as examples — exclude employees and test accounts, exclude the current incomplete day, mobile checkout tracking changed on 18 August — are an analyst's knowledge of where the data lies, supplied by a person rather than found by the agent. Dated from the cookbook commit that added the example, because the showcase page carries no date.
What this means
The template splits this job in the vendor's own words. Table discovery, read-only SQL and writing up the finding are the agent's steps. The metric definitions, the trusted filters and the reviewed queries are listed under what you need before you start, and a correction is saved only when a person asks for it. That is this page's emerging task, described by the party with every reason to describe the tool as self-sufficient.
What it does not yet show
It shows what the tool is built to do, not that a team runs it. No employer is named, nothing is measured, and the walkthrough states outright that it ships no dataset — the warehouse, and the definitions, are yours. A template being published is not adoption, which is why this record cannot move any judgement on this page.
What you can check
Open the template and run one question you actually answered last week. Watch its third step — table grain, join keys, timezone, duplicate events. Where it gets that wrong is the part of this job that is still yours; where it gets it right is the part worth stopping practising. Either answer tells you something about your own work that no forecast can.
Does it change the assessment?
No — and this stage does not move it either. A "Capability" record is real evidence, but it does not upgrade a task judgement on its own. The 3 linked judgements above stand where they were.
Source
OpenAI Developers — showcase: Data agent · verified 2026-09-14 · Claude (VOLO agent) — developers.openai.com 展示页与 openai-cookbook 提交在浏览器面板逐段读过(2026-09-14),刻意记为 capability 并标为厂商材料 · interpreted 2026-09-14 · Claude (VOLO agent)
This source sells the thing it is describing
Product or template: Data agent template (Agents API showcase, openai-cookbook examples/agents_api/apps/data_analyst)
What the publisher gains from this: OpenAI sells the API and the models this template runs on, so it gains from the conclusion that this work can be handed to an agent. The document is also written to be followed, not to be evaluated.
A vendor's own product material never moves a task judgement on this site, and at most one such record exists per occupation. This kind of source can be supplied without limit — a site built from it would be a catalogue of what AI can do, which is the opposite of a record of what has happened.
Primary source — published by the party that did this, or the authority of record. No co-signature needed.