The number of “Forward Deployed Engineer” job adverts has increased significantly by over 700% year-over-year, based on the observations made by Indeed Hiring Lab. Amazon Web Services is investing about $1 billion to establish the Forward Deployed Engineering organization. Microsoft is working to establish the Frontier program, with plans to employ 6,000 staff members through consulting partners. Salesforce has stated its intention to hire 1,000 FDEs. OpenAI also started its own Frontier FDE unit in collaboration with Fortune 500 companies like HP, Intuit, Oracle, State Farm, and Thermo Fisher.
This could not have been accomplished without any reason. It happened due to the fact that the AI pilots hit the wall at the same spot: the demo worked fine and then it ran into the customer’s own systems. Someone has to fill the void and this person is called Forward Deployed Engineer.
In the following article, the evolution of this position will be discussed along with its difference from the position of a solutions engineer and the reason why all businesses are hiring FDEs. Moreover, since just knowing the position does not mean the possibility to create the necessary team, companies use different means of gathering FDEs without any delay.
Where the Role Came From
The label "Hyper-Embedded Engineer" (as it was called in Palantir) by Palantir originated in the 2000s. Palantir's data products were designed for governmental bodies and intelligence agencies in times and places where box selling, installation, and passing the product to a support center was impossible due to highly classified and specific issues of the clients. Therefore, Palantir decided to send its engineers directly to the client(s). The story of Palantir's Hyper-Embedded Engineers from the point of view of time spent working for the client is also interesting because until about 2016, they outnumbered regular engineers.
A model which was developed created a totally new discipline since the FDEs at Palantir are also taught to use failure-analysis frameworks as regards potential problems which are likely to occur while creating the code, because building a bad thing in a working company is more expensive than being idle.
For two decades, the story of Palantir lived on. Then came generative AI, and every enterprise selling AI solutions reached a dead end just like Palantir did with data platforms: a robust, general model may be great, but it makes no sense in a real environment. Open AI built its own Forward Deployed Engineering department. So did Anthropic, and Mistral, and Adobe, and an ever-growing list of businesses that develop AI capabilities internally.
What an FDE Actually Does (That a Solutions Engineer Doesn't)
Accenture has hundreds of Forward Deployed Engineering specialists in open positions.
The common mistake when explaining this position is treating it as a variation of “solution architect” or “sales engineer, ” but that distinction is not superficial. A solution engineer defines the problem, proposes a solution, and passes it to the implementation team. A solutions engineer creates a plan. On the other hand, a Forward Deployed Engineer completely takes charge of the problem: he or she sits in the workflow of the client, identifies a problem that the client wasn’t able to express, comes up with a solution, and monitors its application.
In a recent article for The Batch, Andrew Ng discusses the concept of Field Data Engineering (FDE) as an engineer who works directly inside a client’s organization to fine-tune their workflows according to the client’s requirements. Ng emphasizes that the uniqueness of the engineering role lies not so much in the technical skills but in the soft skills, such as identifying client’s potential needs, prioritizing between multiple requests, explaining the trade-off between alternatives in simple terms, and being able to reject unreasonable requests. This combination of advanced technical ability and the intuition of what should not be built makes the FDE talent extremely difficult to find or train.
The statistics related to hiring also present some interesting findings. For instance, FDE job ads suggest that less than one percent of tasks that an FDE is supposed to carry out can be automated. Most tasks still require human involvement or simply require full human effort. The companies don’t hire FDEs because of the AI potential, they hire them because they are needed in order to use AI in practice.
Build or Buy: The Question Every Ops Leader Is Facing
When a company acknowledges that it needs an FDE function, it has to decide whether to create an internal unit or buy the services of an FDE vendor. Andrew Ng’s analysis lays out the pros and cons nicely: vendor FDEs are really good at making sure that the vendor’s software integrates nicely into your work but when it is impossible to tell which software or model is going to be hot next year, locking yourself into a vendor’s people is a mistake.
An internal FDE team gives you control over your choice of AI stack. It also ensures that your organization knows the workings of its own processes which information is lost when the vendor leaves.
Both of these approaches require developing people. Most corporations already have the raw talent in place: solutions engineers, implementation specialists, and customer-facing service engineers already have the first half of the FDE skill set. What they are lacking is the AI-build part of it and, increasingly, the third part capability to attach a workable prototype to an operational system without disrupting its operations.
Where Training Usually Falls Short and Where It Has to Go Instead
The current AI training modes teach theories, including principles behind the workings of large-scale language models, skills needed for the creation of a good prompt, and the basics of an agent framework. The knowledge gained from these activities may be very helpful, but it is hardly what FDE requires when the product is implemented.
At present, most FDE learning programs assuming such exist do not go beyond the prototype development stage. However, clients run real products rather than prototypes. Thus, there can be different quirks of data accumulated over the years, there can be multiple integrations, which may not be properly documented, and there may be functional users who do not need any changes in the product. So, there is a gap between having a working demonstration and being able to add a functional component without ruining the overall system, where most of the AI projects fail.
How Nuvepro Builds Your FDE Team
Nuvepro has designed the Forward Deployed Engineering program to help fill this gap.
Nuvepro employs Task Intelligence throughout the organization first, which means that all tasks made in one workflow should be classified according to their type using well-known classifications O*NET, APQC and SFIA instead of employing the general-purpose AI training content. Upon the completion of this stage, it gets clear which tasks are considered worthwhile for the establishment of the FDE team and which one should still stay with humans, regardless of the AI development.
Upon completing the previous stage, the new team will move through four levels within a simulated environment.
- Real data domain, systems, and vocabulary introduced into GenAI Sandboxes as opposed to slideware, allowing for unmistakably visible failure in case of giving an incorrect answer.
- Customer Interview Simulator offered to train customer discovery against simulated stakeholders who will only get to the real issue once enough trust has been built. If the solution is pitched prematurely, the dialogue is terminated instantly, just like during a real conversation with a client.
- Prototyping Sandboxes in which the issue identified during the discovery serves as the basis for the build process using managed data and real-life constraints including those set by IT for a good reason.
- Capstone, where the reference product already exists, with all the missing data and workarounds that come with it, and the task is about extending the product without ruining the existing parts of it and showing the outcome to the stakeholder interviewed at the earliest stage.
This capstone has been built by Nuvepro specifically as it is an element of any other FDE curriculum that is overlooked.
The Nuvepro AI Bootcamp: 14 Days to a Task Live in Production
If you are looking to speed up the process of understanding how Ai can be properly utilized by your team, the most efficient way is through Nuvepro's AI Bootcamp. The advantages of joining the Bootcamp include practical tasks offered by a Nuvepro AI Specialist whose task is to give instructions to the whole team and assist in performing the task successfully, thus providing them valuable experience in their work.
There are three versions of the Bootcamp available, at your discretion:
- Pilot – supposed to last for 14 days and to include at most 5 people;
- Sprint – to last for 6-8 weeks and to include up to 15 participants;
- Enterprise when development implementation is performed by implementing the Bootcamp on a custom plan developed for your company.
The Bootcamp’s benefit is that it can be applied by any company that uses any platform, as well as by all participants irrespective of their experience in technology.
The Takeaway
The Forward Deployed Engineer position is more than a trend in hiring that organizations go through according to the need to accommodate response to client’s needs. Therefore it is necessary to find professionals, who will fill a specific vacancy, examine the workflow, and work on its improvement simultaneously. According to the resource mentioned, professionals of such profile are becoming more in demand in the labor market. However, the crucial question lies in the availability of specialists in your organization.
Getting to know the Task Intelligence Platform by Nuvepro is a good beginning to learn where to deploy the Forward Deployed Engineer team. You can get the bootcamp that will include the launching of your first AI–powered project.
You will also benefit from the self-paced sandboxes approach that will allow your team to continue working on other tasks leaving them time for creating the required engineering solutions.