OpenAI demonstrated finance workflows in ChatGPT Work for Financial Services and the Codex desktop app. The session showed how teams could customise workflows with skills and plugins instead of building a separate app.
OpenAI’s applied AI engineering team demonstrated a wealth-management workflow inside an enterprise workspace. It moved from a portfolio change through profit and loss attribution to a client presentation explaining the changes.
ChatGPT Work for Financial Services combines third-party data, models tuned for financial tasks and tools that draft presentations and spreadsheets from templates. Reusable skills guide specific tasks, while plugins bundle skills with MCP connectors that let the model access data and act in outside systems.
For teams building internal AI tools, the session presented a choice between maintaining a custom API-based app and extending the ChatGPT and Codex interfaces. It also highlighted a less visible challenge: delivering fresh financial filings quickly, with context the model can use.
The demonstration leaves production reliability and running costs unproven, so it is not enough to establish a business case.
Teams building internal finance tools have another implementation route to assess before committing to a separate application.
Fresh data and usable context remain engineering requirements even when the workflow sits inside an existing AI interface.
Written automatically by the Made in 10 Newsroom from the source above.
Cohere tests found that a strong lead AI agent paired with cheaper planning agents could match a more expensive team. Using Luna planners under a Sol lead cut costs by 60% without reducing performance against an all-Sol team, but results varied by task.
Iren reported $200 million in quarterly AI revenue and signed a new 5-year cloud contract. The customer prepaid approximately 90% of the associated GPU costs, reducing Iren’s need for additional debt financing.
Physicl has raised ten million dollars in seed funding to make training data for physical AI. Its platform recreates difficult situations in 3D so developers can target weaknesses in their models.