What happened
On June 30, AWS announced a dedicated Forward Deployed Engineering organization backed by a $1 billion investment. Thousands of engineers — many of whom build AWS's own AI services — will embed directly inside customer teams to build production AI systems using the customer's data, governance, and processes.
This isn't a pilot. The NBA, the NFL, Southwest Airlines, Cox Automotive, and Ricoh already have AWS engineers in the building. The largest cloud company on Earth just concluded that selling AI tools isn't enough — someone has to sit inside the business and wire them in.
The detail almost everyone will miss
The number gets the headlines. The engagement model is the story. Read past the $1 billion and look at what AWS says a deployment leaves behind: documented runbooks, a governed knowledge graph living in the customer's own account, and trained internal champions who can run the system without help.
The deliverable isn't software — it's a business that can operate and extend the system after the engineers leave. AWS calls it self-sufficiency by design, and says engagements are structured around business results, not billable hours. That is a direct shot at traditional consulting, from a company that could have simply sold more consulting.
One more detail worth holding onto: AWS is not first. OpenAI and Anthropic both stood up forward-deployed engineering efforts earlier this year. When the three companies with the best view of how AI actually gets used all build the same thing, that's not a product launch — it's a diagnosis.
Why this matters if you run a business
This is a $1 billion confession about where AI actually fails. It doesn't fail in the model. It fails in the gap between the subscription your company already pays for and the workflows where the hours actually go. Enterprise surveys have said it all year — Writer's 2026 study found 79% of organizations struggling to adopt AI even as budgets climb.
If companies with thousand-person engineering departments need experts embedded in the room to get AI into production, a 20-person business shouldn't expect a subscription to move its numbers by itself. The NFL — with AWS engineers working alongside its team — went from start to shipped fan-facing products in weeks. The asset wasn't the model. It was the people who knew how to install it inside real operations.
What to do about it
You don't need AWS's phone number to copy AWS's model. You need to buy AI the way they now sell it. Whether you're hiring help or assigning someone internally, hold the engagement to three tests:
- Embedded, not advisory. The work happens inside your actual tools, with your actual data — not in a slide deck about what you could do someday.
- One real workflow, measured. Start where the hours go — quoting, intake, reporting, follow-up — and let the result argue for the next one.
- Capability stays when the builder leaves. Runbooks, training for the people who own the workflow, and systems running in accounts you control.
The test of any AI engagement is what still works after the person who built it walks out the door. Amazon just made that the standard at the top of the market. It should be your standard too.
