Building an FDE Pod Around Your AI Champion
An FDE pod is a small group that combines embedded engineering with business ownership to deliver a defined workflow. Your AI champion can help it find useful problems and work with users, while the engineer builds and integrates the solution. Clear decision rights and internal knowledge transfer make the arrangement sustainable.
Many organisations already have the beginning of this group. An enthusiastic colleague has collected use cases, run demonstrations and helped others try AI tools. The next stage requires protected engineering time and someone authorised to change the relevant business process. Adding those responsibilities deliberately is more useful than giving the champion another list of projects.
The pod described here is a recommended operating model. Its size and working rhythm should follow the workflow’s needs. A small initiative may need one engineer and part-time business participation, while a complex integration requires additional specialists.
What does the AI champion bring to the pod
The champion understands where people encounter friction and can help the engineer reach the right users. They may also understand why a previous tool was ignored, which unofficial spreadsheets matter or where a proposed change will create extra work.
That knowledge is valuable, but the champion needs a defined remit. Decide whether they represent users, coordinate adoption or own the process itself. Those responsibilities can overlap, yet authority to change a workflow must be explicit.
Give the champion time for the work. If discovery, testing and user support are additional duties squeezed around a full-time role, the engineer will struggle to obtain reliable feedback. The sponsor should decide what existing work is reduced and how participation is recognised.
Who else belongs in an FDE pod
At minimum, identify the person accountable for the business process and the person responsible for engineering delivery. For a larger initiative, a deployment strategist can help translate operational needs into priorities and coordinate stakeholders. Relevant system owners and reviewers join when their decisions are needed.
There is a useful precedent in Palantir’s description of deployment strategists, known internally as Echos, working alongside forward deployed engineers, or Deltas. Palantir describes overlapping roles rather than a rigid split between business and technical work. The transferable idea is to cover both responsibilities and keep them closely connected.
| Responsibility | Suggested owner | Evidence of progress |
|---|---|---|
| Business priority and funding | Executive sponsor | A protected scope and timely decisions |
| Workflow acceptance | Process owner | Agreed measures and acceptance criteria |
| User access and feedback | AI champion or adoption lead | Representative testing and recorded feedback |
| Implementation and integration | FDE | Reviewed, evaluated releases |
| Cross-team coordination | Strategist or named delivery lead | Dependencies resolved with their owners |
| Ongoing maintenance | Internal maintainer or support team | Demonstrated ability to operate the workflow |
One person may cover several responsibilities. Keep the accountability visible even when the team is small, and avoid assigning approvals to someone who lacks the authority to give them.
How do you move from an AI workshop to delivery
Turn the use-case list into a decision about one workflow. Ask which problem has an available owner, accessible information and an outcome the team can observe. Select something narrow enough to investigate without waiting for an organisation-wide programme.
A discovery period should establish what the workflow actually does, which systems it uses and what a useful improvement would look like. The result is a scope the business owner and engineer can both explain, with unresolved dependencies made explicit.
For planning, you might allocate one or two weeks to this discovery and a further four to eight weeks to a bounded implementation. Those are illustrative planning ranges, not a service guarantee. The team should validate them against access, technical complexity and release requirements before committing.
End the initial period with an explicit decision. Continue if the evidence supports the use case, change the scope if a different problem matters more, or stop if the constraints outweigh the value. A workshop creates a starting hypothesis; the pod’s work tests it.
What working rhythm keeps delivery moving
Use a short weekly session to review actual work with the process owner and champion. Examine completed cases, corrections and unresolved decisions. A demonstration is useful when it helps users assess the workflow, rather than simply displaying new features.
Reserve separate time for implementation and technical review. An embedded engineer still needs uninterrupted periods to build and investigate problems. Stakeholder access should make the work clearer without consuming the entire allocation.
Keep the amount of active work small. If several departments request additions, the sponsor and process owner should decide what waits. The engineer should not have to arbitrate competing business priorities alone.
At an agreed monthly review, assess use, operational quality and business value. This is also the time to consider whether the pod has the right capabilities. Repeated delays in one integration may call for a system specialist rather than another generalist.
How do reusable tools reduce dependence on individuals
A pod can deliver quickly and still create a maintenance problem if every workflow uses a separate set of connectors, prompts and deployment practices. Ask which parts should be shared and who owns their upkeep.
In his analysis of forward deployed engineers, Marty Cagan discusses the value of generalising field learning into platform capabilities. For an enterprise team, a practical application is to maintain common integration and evaluation components while keeping business rules and sensitive data specific to each workflow.
Start with reuse that removes demonstrated repetition. Examples include a tested connector, a standard way to record evaluation results or a deployment checklist that reflects your environment. Avoid building a large internal platform before the first workflow establishes what the team needs.
Confirm that the client can use and maintain the necessary components after the engagement. Reuse can lower repeated effort, but unclear licensing or supplier-only operation can introduce another dependency.
How should the internal team learn alongside the FDE
Pair an internal colleague with the engineer on work they will later own. Let that colleague run an evaluation, investigate a failed case and participate in a release. As their understanding grows, have them lead the activity while the FDE observes.
There is current investment in this kind of applied capability. Anthropic announced a $100 million commitment to Claude Frontier Academy in October 2026, aiming to train 10,000 Frontier Deployed Engineers by the end of 2027. Its residency includes engineers leading a real use case at their own organisation. The announcement describes a training objective, not 10,000 already-qualified engineers or proof of customer ROI.
Broader engineering research also points beyond tool access. DORA’s 2025 AI Capabilities Model identifies practices including a clear organisational AI stance, healthy data, version control and small batches. These findings concern AI-assisted software development. They support examining the environment around the engineer, without establishing that a particular pod structure causes better results.
When is the workflow ready for internal handover
Use a practical exercise. Ask the internal owner to run the agreed checks, explain a failed case and follow the recovery instructions. Have them make a small authorised change and assess its effect. This reveals gaps that a documentation sign-off can miss.
Agree what happens after transfer. The internal team may operate the workflow independently, keep occasional advisory support or add delivery capacity through staff augmentation. Document which responsibilities move and which remain with the supplier.
For continuing work, retain a way to evaluate changes and revisit business value. A successful handover should make the next decision easier, whether that means extending the workflow or investing elsewhere.
Rollout IT’s FDE service includes collaboration with internal AI champions and knowledge transfer. A useful scoping conversation starts with the people already involved, the workflow they want to improve and the responsibilities that still need an owner.
Frequently asked questions
Does every FDE engagement need a multi-person pod
No. A limited workflow may suit one engineer working with existing client owners. Add people when the initiative needs distinct expertise or more capacity. Ensure business acceptance, technical delivery and ongoing operation are covered regardless of headcount.
Is an AI champion the same as a deployment strategist
Not necessarily. A champion often helps users explore and adopt AI. A deployment strategist may own broader discovery and stakeholder coordination. Define the responsibilities in your organisation rather than assuming either title includes authority over the process.
Can an FDE pod continue as staff augmentation
Yes, if the ongoing work suits that arrangement and the contract permits it. Make the transition explicit, including who prioritises work, owns outcomes and operates the system. Preserve the knowledge gained during discovery and implementation.
How can we tell whether knowledge transfer has worked
Ask an internal colleague to perform the work they will own. Running evaluations, investigating a failure and completing a controlled change provide stronger evidence than attendance at training alone. Use any difficulties to guide further pairing and documentation.
Sources
Palantir | A Day in the Life of a Deployment Strategist | 8 March 2022
Anthropic | Claude Frontier Academy | 2 October 2026
Google Cloud DORA | Introducing the AI Capabilities Model | 23 September 2025
Marty Cagan at SVPG | Forward Deployed Engineers | 17 September 2025