OpenAI launched ChatGPT Work on Thursday as a professional agent for people who do not write code. It can produce documents, presentations, websites, and other work products, moving the company's pitch from conversation toward finished office output. Reuters described the release as a product launch; its syndication and corroborating coverage presented the same broad set of functions. [1][2][3]

The launch follows two July 8 stories that put OpenAI's ambitions inside a harder ledger. The paper's account of a proposed trillion-dollar IPO against projected losses argued that a sealed filing withheld the revenue and compute detail needed to test the valuation. Its report on a proposed five-percent government stake found no bill, framework, term sheet, or second lab commitment that made the idea binding. ChatGPT Work is more tangible than either proposal: it is a product people can use. It is not yet the missing financial proof.

The gap begins with price. Thursday's reporting established what the agent makes, but not what it will cost, which functions require payment, or what share of users will convert from trying it to buying it. It also did not establish commercial-scale completion rates. A generated slide deck is a product receipt. It is not a revenue line until a customer pays, nor evidence of replacement until the work survives review and reaches its intended use.

That distinction matters because the leap from capability to labor remains unproved. The verified TradingPuzzles post highlighted documents, presentations, and websites, the concrete outputs most likely to resemble office work. The extrapolation is tempting: if software produces the artifact, perhaps it replaces the worker. But the unresolved questions sit between those verbs. Who approves the result? What happens when it is wrong? Which records are retained? What security and audit controls govern company data? The launch coverage does not answer them.

Reuters's product frame is narrower and therefore safer, but it can also hide the consequence. OpenAI is asking future public investors to value a platform rather than a chatbot. A platform needs repeat use, paid conversion, controls that enterprises can accept, and completed work that reduces cost without creating a larger review bill. Reuters did not report any of those measures on launch day. Describing the agent's outputs faithfully is necessary; treating the list as a business model is not.

ChatGPT Work therefore advances the paper's position without overturning it. AI power becomes measurable through instruments: filings, contracts, bills, prices, and controls. This launch supplies one instrument, the working product, while leaving the commercial instruments blank. That is progress, but it is also a more precise test. OpenAI can now be asked not whether its software can draft a document, but whether customers pay for that document and trust the process that produced it.

The next receipt is not another demonstration. It is a price sheet, a conversion figure, an enterprise-control specification, or evidence of completed work at scale. Until one arrives, the product demonstration may invite a replacement story, but Reuters has established only a launch. Investors still lack the bridge between capability and revenue.

-- DAVID CHEN, Beijing