If you have bought AI software this year and felt like it did not quite work out of the box, you were not imagining it. That gap between the demo and the messy reality of your business is exactly why every major AI company is now sending its own engineers to sit inside client companies for months at a time, building the specific system that company actually needs.
This role has a name: forward deployed engineer. Gartner expects more than 85 percent of tech vendors to run this as their main way of delivering AI by the end of 2026. That is a huge shift from the usual software business, where a company buys a license, installs it, and figures out the rest on its own.
The model is not new. Palantir invented it in the mid 2000s because it was selling to intelligence agencies whose problems could not be solved with a standard product. It worked so well that by 2016, Palantir had more of these embedded engineers than regular software engineers, and the approach became central to how the company grew.
What is new is who is copying it now, and how much money is behind it. Microsoft just put 2.5 billion dollars into a unit with 6,000 embedded engineers, already working inside companies like Unilever and Land O'Lakes. AWS has committed 1 billion dollars to the same idea. Google, OpenAI, and Anthropic are all hiring for similar roles, and consulting firms like Accenture and Deloitte are doing the same to avoid losing the work entirely.
Here is the part that matters most for anyone running a business, not just a tech company. Gartner has warned that seven in ten enterprises using this model for their AI projects will eventually have to abandon them. The reason is simple: once the vendor's engineer packs up and leaves, someone inside the company has to keep the system running, fix it when it breaks, and adapt it as the business changes. Many companies do not have that person, and building the AI system in the first place does not automatically create that skill.
This creates a real choice, and it is worth making deliberately rather than by default. You can take the free or discounted embedded engineer a vendor offers, get something built fast, and accept that you may end up paying that vendor a premium fee indefinitely to keep it alive. Or you can push, from day one, for your own staff to sit next to that engineer, learn what they are building, and take ownership of it before the engagement ends.
Insurance companies like Travelers and Liberty Mutual are already doing the second thing, pairing vendor engineers with internal teams so the knowledge stays after the contract ends. That is the smarter move for almost any company without a large technical staff of its own. The vendor's engineer solves this quarter's problem. Your own team's ability to maintain it is what determines whether the AI system is still working a year from now, or sitting unused because nobody knows how to touch it.