FDE and internal engineering: diagnose the bottleneck before allocating work

When an AI project remains in staging, list what prevents release. A running model is progress; quality, permissions, integration, failure handling and operating readiness may still be incomplete. “Code finished” does not establish that adoption is the remaining problem.

FDE and internal engineers can both build, evaluate and operate systems. Compare actual skills, allocated time and decision access rather than assuming internal teams only maintain services or outside teams automatically drive change. Three questions help separate the bottlenecks.

What work and evidence are missing?

Check main tasks and exceptions, available data and interfaces, and agreed permissions. Record unresolved errors, dependencies and missing acceptance evidence. Turn “not ready” into items a named person can complete or decide.

GAO's AI accountability framework covers governance, data, performance and monitoring. It helps identify overlooked work, not prescribe a fixed split between modeling and adoption or a natural role for one team type.

Suppose a hypothetical quality-document query tool answers development examples but has not been tested on obsolete documents, cross-department permissions or missing sources. Technical and evaluation work remain. List them as pilot prerequisites with delivery and acceptance owners rather than diagnosing user resistance.

Do available skills come with allocated time?

Record skills separately from assigned capacity. Internal teams may have relevant expertise without available time; external teams also face staffing and delivery constraints. A proposal's headcount does not replace named participants, allocated work and support arrangements.

Address what blocks the next step; purchasing a complete delivery model is not always necessary. A defined integration or testing task can be compared through its output and acceptance evidence.

Who can make process and access decisions?

An engineer proposing a solution does not necessarily have authority to change a workflow, expose sensitive data or accept risk. Name the decision-maker, required evidence and response process. Outside presence grants no automatic authority; internal teams may already have the necessary organizational support.

When access is restricted, evaluate approved samples, controlled environments, remote collaboration or reduced scope. If these cannot support evaluation, record the dependency and adjust the next stage rather than treating access to all raw data as an FDE requirement.

Thoughtworks' enterprise AI services describe work spanning engineering and transformation. Confirm the particular skills, time and responsibilities in the actual proposal.

Choose an executable allocation

Internal delivery requires capability, capacity and decision support. External delivery requires approved access, acceptable scope and receiving roles. Shared work needs connected delivery, review and dependency ownership so neither side waits indefinitely for the other.

For the hypothetical query tool, internal staff might approve sources and access, external engineers deliver specified integration and tests, and both review pilot evidence. Assess this arrangement against actual people, costs and constraints; it is not inherently cheaper or faster.

Before handover, have receivers perform necessary deployment, configuration, testing or incident tasks. Check documentation, code and rights against contracts. Departure does not establish independent operation. Assess adoption, quality, permissions and operational acceptance separately rather than declaring completion from daily use.

Use the existing FDE/in-house staffing comparison for options and handover capability, or the delivery method for first-stage and operating ownership. Bring bottlenecks and pending decisions before buying additional work.