What Formula 1 explains about the OpenAI and Oracle partnership

Originally published on NewHack, on November 4, 2026.

In André Lopes's keynote in São Paulo, the pit crew became a metaphor for the work that makes AI work after the demo.

You can have the best car, one of the best drivers, and a track in ideal conditions. Even so, when the car pulls into the pits, the race depends on what the team can do there.

André Lopes, from OpenAI, used this image at Oracle AI Experience Brasil, in São Paulo, to explain the premise of the partnership with Oracle. The two companies want to act as that pit crew, helping to choose where to apply AI, design the solution, follow the rollout, and adjust whatever is needed once it starts running.

That was the part of the keynote that caught my attention the most. In the analysis I published after Oracle AI World, I wrote about ChatGPT as a gateway into enterprise systems. In São Paulo, the conversation moved forward to what happens after that door opens. Someone needs to know the process, know what to measure, and be able to step in when the operation strays from what was expected.

Before you accelerate, time the lap

The case presented came from Oracle itself. The recruiting team used to take two to four days to prepare a market survey. According to the account published by OpenAI, with ChatGPT Work that preparation now takes 15 to 20 minutes. The gain refers to the survey, a specific step in recruiting. The team already knew its starting time and knew exactly which task it wanted to speed up. Oracle case published by OpenAI.

The question André left for the audience was which process each company would choose to start with. It sounds simple, until someone has to say what result they expect, who is accountable for it, and what evidence will justify the next step.

The work that shows up when the car enters the pits

André drew another important distinction. A process can have worked for years for an experienced team and still depend on decisions that no one ever wrote down. People know which spreadsheet prevails when the numbers diverge, which exceptions require a phone call, and who has the authority to approve something outside the routine.

Imagine a procurement team trying to understand why an order became more expensive. The AI can gather the order, the budget, and the contract. But if there are two versions of the contract, which one counts? If the price adjustment falls within tolerance, who can move forward? When information is missing, who gets the open item?

That's where the pit crew's role becomes concrete. Someone needs to spell out these rules, test what happens in exceptions, and follow the flow after rollout. Every case sent for review can show where a piece of data is missing, where a rule turned out ambiguous, or where human judgment still needs to carry on.

The Oracle AI Workspace, still under development, aims to bring Oracle applications and data closer to ChatGPT, following the organization's permissions and approvals. Oracle App Composer plans to use Codex and other AI tools to build applications tied to approved data and workflows. An investigation that starts as a conversation can, if it recurs often enough, turn into an internal application with clear steps and someone responsible for each exception. OpenAI announcement.

How I would choose the first lap

I would start with a task that happens every week and whose outcome someone can review. I would run the AI in parallel with the current process, without altering records or approving requests. I would compare the time to delivery, including review, and watch which exceptions got routed correctly. If the AI produces a draft in minutes and the team spends an hour fixing the sources, that hour counts too.

For an AI startup, this is an opportunity to show value with precision. When the customer can point to which work now gets delivered faster, how much effort still stays with the team, and how exceptions get resolved, they have evidence to expand usage. The first lap measures what the car can do. The ones that follow show whether the team can repeat the result under real conditions.