Amazon Web Services launched a new program on Tuesday that sends teams of its own engineers to live and work inside client companies for around 45 days at a time, building AI systems directly into the client's operations. The program, called AWS Forward Deployed Engineering, is backed by $1 billion in internal resources and is already active with clients including Southwest Airlines, Cox Automotive, the NFL, and the NBA.
The model is simple to understand. Instead of selling software and leaving the client to figure out installation, AWS sends a team of roughly five to six engineers who embed with the client's staff, connect to the client's internal data, build working systems, and then hand everything over when the engagement ends. The client keeps the software, the documentation, and enough internal knowledge to operate independently.
This is not an isolated move by Amazon. OpenAI launched a $4 billion venture called The Deployment Company in May, backed by 19 investors including TPG and Bain Capital. Anthropic launched a $1.5 billion joint venture the same week, with Blackstone and Goldman Sachs as founding partners. All three landed on the same answer within two months.
The reason is a hard reality that has been building for years. Around 70 percent of AI pilots inside large organisations never reach production, meaning they are tested, demonstrated, and then quietly abandoned. The model that made AI look impressive in a boardroom kept failing once it met real data, real security teams, and real internal politics. Embedding engineers directly inside the client is a structural answer to that failure rate.
The original version of this model came from Palantir, the defence and intelligence software company, which pioneered the approach more than a decade ago. Palantir's stock returned roughly 640 percent over five years once that model took hold at scale. The AI labs have noticed.
There is a meaningful difference between what AWS is doing and what OpenAI and Anthropic set up. The OpenAI and Anthropic ventures are formal joint ventures with private equity firms, which give the labs access to thousands of portfolio companies through their PE partners' existing client networks. AWS is funding the program from its own balance sheet, which means it keeps full control but also carries the full cost.
For business operators, this matters in two ways. First, if your company runs on AWS and has been stuck on an AI project that has not moved to production, this program is worth asking your account manager about directly. The pitch is a dedicated team, a fixed timeline, and software that runs in your own environment when they leave.
Second, the broader pattern tells you something about what is coming from AI suppliers generally. The era of being sold a software licence and left to integrate it yourself is ending. The companies building AI are now competing on whether they can make it actually work inside your organisation, not just whether their model scores well on a test. That changes what a good AI contract looks like, and it changes which suppliers are worth talking to.