AWS just committed a billion dollars to putting engineers inside its customers. The model was never the hard part.

AWS has put a billion dollars into embedding its own engineers inside customer businesses, following OpenAI and Anthropic. When the people who build the models spend that kind of money on people, it tells you where the hard part really is.

Three of the biggest names in AI have, in the space of a few months, quietly agreed on something. The model in the box is not the product.

Last Tuesday, AWS announced a dedicated Forward Deployed Engineering organisation, backed by a billion dollars, to embed its own engineers inside customer teams and build agentic systems in place. It follows OpenAI and Anthropic , who have each stood up their own versions in recent months (reportedly valued at four billion and one and a half billion). The approach was pioneered by Palantir. It is now the fashionable answer to a very unfashionable problem: this stuff is genuinely hard to get working inside a real business.

Read that back. The companies that build the models are spending billions of dollars on people. Not on bigger models, on people who sit next to your team and wire the thing into how you actually work.

The tell is in the spend

For two years the industry argued about which model was best. Benchmarks, context windows, leaderboards. The labs have now told us, with their wallets, what they think the hard part is. It isn't the model. It's the messy, specific, human work of getting a model to do something useful with your data, your processes and your governance.

That is the whole reason "forward deployed" exists. You take an engineer who understands the technology, you put them inside the business for a while, and they close the gap between what the tool does in a demo and what it does in your operation on a Tuesday. AWS is honest about the cost of that. It means keeping a standing corps of expensive engineers on the road. They are doing it anyway, because the alternative (ship the tool, wish the customer luck) has not worked.

You are not getting an AWS engineer

Here is the part that matters for the rest of us. The customers named in the AWS announcement are Southwest Airlines, the NFL, Cox Automotive, Ricoh. Those embedded engineers are not coming to a hundred-person business in Parramatta or Penrose. If you run a mid-market organisation in Australia or New Zealand, the billion dollars is not for you.

But the lesson is. The gap those billions are aimed at is not a model gap, and it is not a gap you can close by buying another licence. It's an implementation and capability gap, and it is the same shape in a fifty-person firm as it is at Southwest. Someone has to sit between the business and the technology and make it real. That role is now scarce enough that Amazon, OpenAI and Anthropic are all bidding for it.

The line I would underline

The most important sentence in the AWS announcement is not about speed or scale. It is this:

Customers leave AWS FDE deployments with both new solutions and new engineering capabilities ... skills, workflows, and patterns they can use to innovate independently.

Self-sufficiency, designed in. Their own framing is that customer engineers move "from observers to co-builders to autonomous operators". That is the right test for any help you bring in, whether it is a global vendor or someone like us. Does the engagement leave capability behind, or does it leave a dependency? A deployment that only works while the expert is in the room is not a win. It is a subscription to somebody else's expertise, and you will be renewing it forever.

I would be a little sceptical of the speed pitch. AWS talks about compressing months into days, and vendors always talk about speed. But the structure underneath is right, and it is the bit worth copying at any size: bring the expertise close to the work, build in the open with your own people, and measure success by what your team can still do after everyone goes home.

The uncomfortable read for anyone who has spent two years shopping for the best model: the model was never the hard part. The hard part was always going to be your organisation. That was true before AWS spent a billion dollars proving it, and it will be true long after.