The next generation of Airmen and Guardians won't just use AI tools, they will write the software that wraps AI into mission systems. Software factories, cyber units, and acquisition development teams need an upskilling model that treats AI as a first-class part of the engineering craft, not a separate specialty bolted onto a developer's job description. This session presents a practical, code-centric upskilling program that Insight has refined across our own delivery teams and large public- and private-sector enablement engagements, translated into a roadmap the DAF can adopt today.

The program is built around four engineering competencies, each tied to working code, not slideware.

1. AI-Augmented Development. Every developer should be productive with an AI pair-programmer in their daily flow and just as importantly, should know when not to trust it. The session will cover prompt patterns for code generation, code review with AI assistance, AI-assisted refactoring, and the disciplined habits (small diffs, strong tests, code review of AI output) that keep velocity gains from becoming defect debt. Learners practice on real legacy code drawn from sanitized DAF-style scenarios.

2. Building With AI APIs. Most mission AI features today are calls to a foundation model wrapped in domain logic. The session will teach engineers how to build production-grade applications that integrate AI capabilities cleanly: structured prompts, function calling and tool use, retrieval over authoritative documents (so the model answers from doctrine instead of guessing), guardrails and content filtering, fallbacks and graceful degradation, observability for AI calls, and cost and latency budgeting. The education from this phase is a working mission-style microservice, such as an after-action-report drafter, a logistics-question assistant, a cyber-alert triage helper, all built, tested, and deployed with the aid of AI.

3. Engineering Quality for AI Features. AI features fail differently than traditional code. We will cover test strategies that actually work for non-deterministic systems: golden-output regression suites, behavior-driven evaluations, red-team prompt libraries to catch prompt-injection issues, and CI pipelines that fail a build when model behavior drifts. Attendees will see how to stand up a full pipeline that gates AI feature releases on the same kind of objective evidence the DAF demands of any safety-relevant change.

4. Secure, Mission-Ready Delivery. AI features live inside the same DevSecOps reality as the rest of the codebase, STIG-hardened base images, signed builds, SBOMs, and ATO-aligned evidence. We will walk engineers through deploying AI-enabled microservices through a Platform One–style pipeline, including the security review patterns that matter most for AI: data handling, model and prompt provenance, and logging that supports incident response.

We will share the playbook from Insight's own internal rollout as well that includes an actionable 90-day starter plan for a USAF software factory or cyber unit ready to begin.

Key takeaways:

• A code-centric, four-competency curriculum that turns developers into AI-native engineers
• Practical patterns for AI pair-programming that protect quality and velocity
• A production blueprint for building, testing, and securing AI-enabled mission features
• Test and CI patterns that keep non-deterministic AI features inside engineering rigor