Leading the architecture effort for LERETA’s $20M legacy modernization program, working across a team of 30+ developers over multiple years, the challenge was not identifying the right technology. The technology options were clear enough. The challenge was governance: ensuring that the decisions made at the team level accumulated into an architecture the board could fund with confidence and the CTO could defend. The wall-sized architecture diagrams that became known internally as the Livermore Report were not about aesthetics. They were about making the full scope of the technical reality visible — what existed, what needed to change, and in what sequence — so consequential decisions could be made by people with accountability for the outcome.
That governance challenge is exactly what enterprise vibe coding is surfacing now, at higher speed and higher stakes. AI coding tools produce working code faster than the organizational structures designed to review and govern it. A 2026 CIO analysis of enterprise vibe coding and legacy modernization describes systems capable of reading legacy code, extracting business rules, and generating modern replacements at scale. The capability is real. What matters is the governance gap between what the tools can generate and what enterprises can responsibly ship.
stateDiagram-v2 direction TB state "AI Tool Enabled for Team" as Enable state "Governance Framework Established" as Gov state "Code Generated with Tool" as Gen state "Architecture Review Gate" as Gate state "Security and Compliance Check" as Sec state "Production Deployment" as Prod state "Ungoverned Accumulation" as Haunt [*] --> Enable Enable --> Gov: governance first Enable --> Gen: tool first Gov --> Gen Gen --> Gate Gate --> Sec: meets standards Gate --> Gen: revision required Sec --> Prod Prod --> [*] Gen --> Haunt: no review gate Haunt --> [*]: technical debt compounds
What Enterprise Vibe Coding Actually Involves
The demo version of vibe coding is a developer describing a feature in natural language and watching a capable AI tool produce working code. That works. The enterprise version of that interaction runs through a set of gates that have no equivalent in the solo developer context.
IP and data governance. What information is the developer sharing with the AI tool? Customer data, proprietary business logic, and regulated information have to stay out of third-party AI tool contexts. Enterprises that have not established a policy on this are operating with developers making individual judgment calls on a question that has regulatory and legal implications.
Starter templates and architecture constraints. Enterprise teams using vibe coding tools without approved starter templates are generating code that reflects the tool’s defaults, not the organization’s standards. As a practitioner-maintained documentation of enterprise vibe coding governance gaps maps in detail, shadow AI use, comprehension debt, and what practitioners call haunted codebases — code that runs but that nobody fully understands — are the direct result of scale without structure.
Code review standards for AI-generated output. AI tools generate plausible code that can pass a cursory review and fail a compliance audit. Review standards designed for human-written code need to be extended explicitly to AI-generated code, because the failure modes differ. A human developer who misunderstands a requirement writes something that reflects that misunderstanding. An AI tool that misunderstands a requirement writes something that looks authoritative.
Ownership and accountability. Who owns AI-generated code when something goes wrong? This question sounds abstract until something goes wrong. The answer has to exist before the first production deployment.
The Modernization Intersection
Enterprise vibe coding and legacy modernization are converging. Organizations with decades-old systems are exploring whether AI tools can accelerate the extraction of business rules and the generation of modern replacements — and to a meaningful degree, they can. But the organizations that extract durable value from this capability are the ones with the governance layer in place before the generation begins. Without that layer, the new AI-generated system inherits the problems of the legacy system — now wrapped in modern syntax, still governed by nobody.
What Governance Looks Like in Practice
At LERETA, making the full technical reality visible was what unlocked board confidence and the funding to act on it. The architecture documentation was not bureaucracy — it was the basis on which consequential decisions could be made by people accountable for the outcome. Enterprise vibe coding governance is the same function in a different form: making the technical reality of AI-generated code visible enough that the decisions about what to ship, review, and maintain can be made with appropriate oversight.
That is not a reason to avoid AI coding tools in enterprise settings. It is the condition under which they produce durable value rather than technical debt at AI speed.
