FDE Council Professional Capability Framework
Draft v0.3 is the current open framework. It keeps the 11 mandatory core domains. When a deployment uses models or agents, AI and agent engineering says what the FDE must do. That work is required in that context. It is not a twelfth core domain.
Published by FDE Council · Draft v0.3 · · Updated
This is a draft
This is the current draft for community review. It has not been validated by a job task analysis, it is not a certification standard, and nobody can be assessed or certified against it. Earlier drafts stay at v0.2 and v0.1. The free self-assessment scores these 11 domains. It does not score AI and agent engineering as a separate domain.
New to the role? Start with what a Forward Deployed Engineer is.
What changed from v0.2
The 11 core domains are the same domains, in the same order. Seven descriptions now say explicitly what v0.2 left implicit. The other four descriptions are unchanged. Layers 2 and 3, which v0.2 named but did not fill, now have their first entries. Behavioral indicators are still not in this draft. The detailed indicators remain in v0.1.
- Domain 1 names who decides, and a go, reshape, or stop recommendation when data, access, sponsorship, or feasibility is missing.
- Domain 3 names the choice to build, configure, or extend the product, including when no new software is the right answer.
- Domain 6 keeps one test for every solution. When the system uses a model, the evidence also includes an agreed test set, allowance for answers that are not deterministic, and a regression check after the model changes.
- Domain 7 names runbooks, knowledge transfer, support, and a named owner at exit. When the system calls a model, operating cost includes token and inference cost, and the runbook names the model version.
- Domain 8 names responsible-use controls: transparency, a record of how decisions were made, and human review where a person must be able to intervene.
- Domain 9 names communication with executives, operators, and engineers, and support for resistance and for changes to roles and workflows.
- Domain 10 names when to narrow or stop, and enough of the commercial model to judge cost against value.
Context-required. AI and agent engineering is separate from the domain edits above. It names language models, prompting, structured outputs, retrieval, tool calling, and multi-step agents. It also names a person in the loop where the risk requires it, mitigation of prompt injection, controls for data leaving through tools, model version drift, and a plan for a provider retiring the model.
Case studies, account growth, and travel are not core capabilities. Expansion sits with account management, which is a different role.
Structure
The framework has three layers:
- 11 mandatory core domains. Every FDE should develop capability in all eleven.
- Context-required capabilities, used when a deployment calls for them. They do not replace a core domain.
- Advanced specializations that build on the core. They are not required of every FDE.
The core domains are competency domains, not sequential project phases. An FDE draws on several of them at once, and returns to them throughout a deployment.
The expected independence and depth in each domain increase with experience. Three indicative levels describe that progression:
| Level | Typical scope (indicative) |
|---|---|
| Associate | Delivers well-defined parts of a deployment under guidance. Reliable on scoped tasks; asks for help early. |
| Practitioner | Owns a deployment end to end for one customer or workstream, working independently in ambiguous environments. |
| Lead | Owns several or high-stakes deployments, sets the approach for others, shapes the vendor’s product and playbooks, and is trusted by customer executives. |
Years of experience are deliberately not used to define levels. Trust with customer executives is a description of Lead scope, not a separate skill. The skill of adapting communication for executives sits in Domain 9.
The 11 mandatory core domains
Every FDE should develop capability in all eleven. Descriptions marked “unchanged” are the v0.2 wording.
| Domain | What an FDE must be able to do |
|---|---|
| 1 Discovery and Problem Framing | Enter an unfamiliar environment, learn the domain, observe actual work, distinguish symptoms from root causes, identify who decides, and name the problem worth solving. Recommend go, reshape, or stop when data, access, sponsorship, or feasibility is missing. |
| 2 Workflow Design and Value Definition | Design an improved way of working, including user experience, human decisions, exception handling, measurable benefits, and conditions for success. Unchanged from v0.2. |
| 3 Solution Architecture and Enterprise Integration | Design a practical solution within existing systems and constraints. Choose whether to build, configure, or extend the product, including when no new software is the right answer. Make defensible architecture choices and diagnose problems across system boundaries. |
| 4 Data Foundations and Engineering | Understand what data means, establish permitted access, assess quality, and build dependable data models, transformations, and pipelines. Unchanged from v0.2. |
| 5 Software Engineering and Rapid Experimentation | Read and modify unfamiliar code, build working solutions, test risky assumptions quickly, debug effectively, and maintain appropriate engineering quality. Unchanged from v0.2. |
| 6 Verification, Evaluation and Outcome Measurement | Establish whether the solution works, how it fails, whether users can use it, and whether it improves the target workflow, with evidence appropriate to the risk. When the system uses a model, include an agreed test set, allowance for answers that are not deterministic, and a regression check after the model changes. The general test applies to all FDE work, including systems that do not use AI. |
| 7 Production Deployment, Reliability and Operations | Deploy safely, monitor performance, recover from failures, and manage operating costs. When the system calls a model, operating cost includes token and inference cost, and the runbook names the model version, watches for version drift, and says what replaces the model if the provider retires it. Leave the customer able to operate the system: runbooks, knowledge transfer, support, and a named owner. |
| 8 Security, Privacy, Safety and Governance | Identify risks, implement appropriate controls, respect authorization and data boundaries, secure required approvals, and escalate work beyond personal competence or authority. Apply proportionate responsible-use controls: transparency, a record of how decisions were made, and human review where a person must be able to intervene. |
| 9 Embedded Collaboration, Trust and Adoption | Work alongside customer teams. Adapt communication for executives, operators, and engineers. Handle disagreement and resistance, support changes to roles and workflows, build the customer’s capability, and help the solution become part of normal operations. |
| 10 Delivery and Commercial Judgment | Manage scope, priorities, dependencies, uncertainty, costs, and commitments. Make sound delivery trade-offs without overpromising or bypassing controls. Know when to narrow or stop, and understand the commercial model well enough to judge cost against value. |
| 11 Field Learning, Product Feedback and Reuse | Convert field experience into actionable feedback, reusable assets, and improvements to future deployments while respecting confidentiality, ownership, and product boundaries. Unchanged from v0.2. |
Design, build, verify, operate
Domains 2, 5, 6 and 7 answer four different questions:
- Design (Domain 2): What should improve, and how should the workflow operate?
- Build (Domain 5): Can we construct the proposed solution?
- Verify (Domain 6): What evidence shows it works and delivers the intended benefit?
- Operate (Domain 7): Can it continue working safely and reliably after the FDE leaves?
Evaluation is a requirement for all FDE work, including systems that do not use AI. When the system uses a model, Domain 6 also includes an agreed test set, allowance for answers that are not deterministic, and a regression check after the model changes. For a practical way to judge a live deployment, read how to measure whether an FDE deployment succeeded.
Context-required capabilities
Use a context-required capability when the deployment calls for it. Skip it when it does not. These capabilities sit on top of the core domains. They do not replace Domain 5, Domain 6, or Domain 8.
AI and agent engineering
An agent, in this draft, is software that takes a goal and carries it out over more than one step, usually by calling tools or a model. A language model is in this same capability when the deployment uses one, including prompting and structured outputs. Many FDE deployments never use either. When a deployment does, the FDE has to be able to build or integrate it.
That means six things:
- Prompting and structured outputs, when a language model has to return something the rest of the system can use.
- Retrieval over the customer’s own content, when the agent needs facts the model was not trained on.
- Tool calling, including a multi-step agent that chooses the next tool from what the last step returned.
- A person who can intervene where the risk requires it, before the agent acts on a customer system or a customer decision.
- Mitigate prompt injection. Control customer data leaving through a tool. Prompt injection cannot be fully prevented.
- The model version in production: watch for version drift, and say what replaces the model if the provider retires it.
The surrounding engineering stays in Domain 5. Whether it works, and whether the workflow improved, stays in Domain 6. For a model, that includes an agreed test set, allowance for answers that are not deterministic, and a regression check after the model changes. Token and inference cost stays in Domain 7. Security, privacy, and approvals stay in Domain 8. The row below is this same capability.
| Capability | When it applies | What an FDE must be able to do |
|---|---|---|
| AI and agent engineering | The deployment uses models, retrieval, tool calling, or multi-step agents. | Build or integrate the model or agent: prompting, structured outputs, retrieval over customer content, tool calling, and multi-step workflows. Keep a person able to intervene where the risk requires it. Mitigate prompt injection, and control customer data leaving through a tool. Name the model version, watch for version drift, and plan for the provider retiring the model. Whether it works remains Domain 6, including an agreed test set, allowance for non-deterministic answers, and a regression check after a model change. Token and inference cost remains Domain 7. |
| Restricted and air-gapped environments | The customer limits installation, network access, or ongoing operation, including isolation from the public internet. | Deploy, monitor, and hand over the solution inside those limits, without a path that assumes open network access. |
| Regulated-data constraints | The data brings residency, consent, retention, or sector rules beyond ordinary access control. | Work inside those rules, say which rule applies, and escalate what is outside personal authority. Ordinary access control remains Domain 4 and Domain 8. |
Advanced specializations
An advanced specialization builds on the core. It never replaces a core domain. v0.3 names two examples so the layer is no longer only a label. It is not a closed list, and it is not required of every FDE.
| Specialization | Builds on | What it adds |
|---|---|---|
| Portfolio practice leadership | Domains 9, 10, and 11, at Lead scope | Owns several deployments, sets methods other FDEs reuse, and turns repeated field lessons into product and playbook changes. |
| AI evaluation lead | Domain 6 and AI and agent engineering, at Lead scope | Owns how model-based systems are tested: the agreed test set, allowance for answers that vary, and the regression check when a model changes. Not required on deployments that do not use a model. |
Industry-specific tracks are not in this draft. Add one through the feedback form if a deployment context needs it often enough to name.
Design principles
- Vendor-neutral
- Outcome-anchored
- Observable and assessable
- Built to be ready for ISO/IEC 17024 later (this draft is a starting hypothesis for the job task analysis, not a substitute for it)
- Standard-setting kept separate from training, including any training affiliated with FDE Council’s founder. The founder’s own training is held to the same rule: it may say “aligned to the open framework” and nothing stronger.
- Specialization is additive: advanced specializations build on the core domains and never replace them
- Open by default
Validation roadmap
- Community review of the draft (open comment period).
- Job task analysis: a structured survey of practicing FDEs and FDE leaders plus subject-matter-expert workshops. (State of FDE 2026 gives early input; it is not a substitute.)
- v0.x to v1.0: revise, and publish the job-task-analysis method and results.
- Later and separate: any credential scheme, under a separately governed committee.
How to give feedback
Name the domain number or the context-required capability, say what is missing, wrong, or unclear, suggest wording, and tell us your context (company type, region, level). Please do not share confidential customer information. If you are mapping a learning path to this draft, read how to map a learning path.
Versioning and license
Releases use MAJOR.MINOR. 0.x releases are drafts; 1.0 will be the first release validated by a job task analysis. Earlier versions stay available and are not edited when a later draft is published.
FDE Council Professional Capability Framework v0.3 © 2026 FDE Council, licensed under CC BY 4.0. You may share and adapt it for any purpose, including commercially, with credit. Use does not imply endorsement by FDE Council.
Editor: Noor Surani. Reviewers will be listed as the advisory board forms.
How to cite this framework
FDE Council. (2026). FDE Council Professional Capability Framework (Version 0.3, draft). https://www.fdecouncil.com/framework/v0.3/. Licensed under CC BY 4.0.
To cite a single domain, give its number as well, for example “FDE Council Professional Capability Framework v0.3, Domain 6.” Earlier drafts: v0.2 and v0.1.
Change log
- v0.3November 2026Current draft. Same 11 core domains. Seven descriptions made explicit, including model evaluation in Domain 6 and token and inference cost in Domain 7. Context-required capabilities: AI and agent engineering, restricted environments, and regulated-data constraints. AI and agent engineering mitigates prompt injection and names model version drift. Two example advanced specializations, including AI evaluation lead. The v0.1 and v0.2 texts were not rewritten.
- v0.2October 2026Earlier draft. 11 mandatory core domains. Evaluation is its own domain and applies to all FDE work.
- v0.1October 2026First public review draft: 7 domains with competencies and behavioral indicators at 3 levels.