Most companies have paid for AI and see little of it in their margins. The money gets stuck between the pilot and daily use: licenses and demos don't change how a process runs. What Uber did this month points to what does: engineers sitting with the teams that do the work, two weeks at a time, on one part of the job.
What Apollo's Chief Economist Said
On October 24, Torsten Slok, chief economist at Apollo Global Management, wrote that AI has delivered "zero financial payoff to corporate America outside of tech." He looked at the "S&P 493", the S&P 500 without Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla. In that group, he found, profit margins and revenue per worker appear flat.
Read it as one economist's view, not a measured result. The report gives no margin or revenue-per-employee figures, and Apollo, a private equity firm, has its own reasons to comment on where returns come from. His explanation is worth keeping, though: "AI may be a great calculator, but in business, intelligence is mostly judgment."
It also matches what many CEOs and CFOs see in their own numbers as they set 2027 budgets. AI spending went up. The cost of running the business didn't come down.
Where the Money Gets Stuck
Rarely in the model. In most companies, it sits in one of three places:
- Pilots that stay demos. The pilot works on clean sample data, then meets the order that arrives as a PDF attached to a forwarded email, and never goes further.
- Licenses nobody builds on. Every employee gets an AI assistant. A few write better emails. The process that eats the most hours stays exactly as it was.
- No one owns adoption. A vendor hands over the software and leaves. Three months later, the team is back in the spreadsheet.
In each case, the process is the same on Monday as it was before the project. Margins move when a process changes: fewer manual touches per invoice, fewer orders re-keyed from email into the enterprise resource planning (ERP) system, faster offer approval.
What Uber Did Differently
Two days earlier, on October 22, Uber said it would put its roughly 30,000 employees on ChatGPT Work (TechCrunch). The license is the headline. The more useful part is how Uber got there.
Sumanth Sukumar, Uber's VP of engineering, said the company "paired 30 AI engineers with teams across Uber and gave them two weeks to automate key parts of their job." An earlier pilot covered 1,000 employees in finance, operations, product and other departments, and about 80% of the early adopters weren't engineers. Uber also said what it will measure next: "how employees are using the technology, the kinds of workflows they develop and how it helps them work more effectively."
Two caveats. Uber didn't say which tasks the pods automated, or whether the pods were part of the pilot. And Uber has thousands of engineers; a 500-person distributor can't pull 30 of them off its roadmap. The lesson is the method, not the scale.
Why the Method Works
Uber's pods have three things most stalled AI projects lack:
- Engineers sit with the people who do the work. They see the WhatsApp orders, the spreadsheet that bridges the ERP and accounting, and the exceptions nobody wrote down.
- A short, fixed time box. Two weeks forces the team to pick one small process instead of a transformation roadmap.
- Success is measured in workflows, not log-ins. The question is whether the work runs differently, not whether people opened the tool.
This is how the forward-deployed engineer (FDE) model works: an engineer embeds with a team, learns how the process runs today and builds automation into the systems the team already uses. For the full definition, see What Is a Forward-Deployed Engineer?
What a Mid-Market Company Can Do in the Next Two Weeks
You don't need 30 AI engineers to use the method. You need one process and one engineer who stays close to it.
- Pick one small process people do by hand, in volume: order entry from email, invoice matching, carrier quotes, loan document checks.
- Write down today's baseline: time per order or invoice, error rate, how many people touch it.
- Put an engineer next to the team that runs it, not in a separate project room.
- Ask for a working prototype on real or anonymized data.
- Decide the next step by use, not by the demo: is the team doing the work differently?
Start with one process. Measure it. Keep what the team uses.
Want to Discuss a Project?
At Cody Solutions, forward-deployed engineers build a working prototype for one small process in two weeks or less, and the first small automation is free. After that, every process has a fixed price.
We only count an automation as done when your team uses it every day. If yours isn’t using it after launch, we’ll put that budget toward automating a different process at no extra cost.
Let's talk. Click on a button on top of the article to get in touch.
