Agentic Engineering And Enterprise Architecture Discipline

Narrative

The strongest practitioner signal is that "vibe coding" is not the end state; it is a weak, prototype-scale label that serious teams are replacing with spec-driven, harnessed, and governed agentic engineering. DORA’s 2025 research treats AI as an amplifier of organizational quality, while its 2026 follow-up explicitly describes a tension between faster code creation and more auditing, verification, and downstream instability. ThoughtWorks and Martin Fowler contributors sharpen that into concrete engineering practices: reference applications, context engineering, harness engineering, human-on-the-loop workflows, and explicit attention to internal quality, supply-chain risk, and long-term maintainability. InfoQ mirrors the same shift in mainstream practitioner reporting, with Kiro, Dapr Agents, and related articles emphasizing acceptance criteria, design docs, retries, observability, and Kubernetes-native control as the differentiators between demos and production.

Across leadership and platform-adjacent sources, the pattern is consistent: enterprise adoption is moving from individual productivity claims to system-level governance, reliability, and cost control. MIT Sloan and HBR focus on management rules, auditing, and cross-functional operating models; CNCF shows the cloud-native substrate hardening around Kubernetes, conformance, observability, and production agent frameworks; Stack Overflow’s survey data shows high usage but weak trust, with most developers still not using vibe coding in professional work and many spending more time fixing almost-right output. The empirical thread is not that agents remove software engineering disciplines, but that they make existing disciplines non-optional: architecture boundaries, secure SDLC, testing discipline, telemetry, incident response, and governance become the mechanism by which agentic speed can be converted into durable enterprise value.