From User Experience To Vibe-Coded Experience: Why Governance Matters
Shailesh Manjrekar is the Chief AI and Marketing Officer at Fabrix.ai, inventor of “The Agentic AI Operational Intelligence Platform.”
gettyFor two decades, enterprise software has been organized around user experience. Designers decided what an operator would see, engineers built it, and users adapted. SREs, network engineers and security analysts lived inside dashboards someone else imagined, and filed a ticket when those dashboards fell short.
That model is beginning to break. Vibe coding lets people describe intent in plain language and get back a working dashboard, workflow or agent. In an HFS Research survey of 100 U.K. and Ireland business and IT leaders, conducted with Cognizant, respondents expected 44% of knowledge workers to be actively vibe coding within 12 months. The result could be a shift from user experience (UX) to a vibe-coded experience (VX), where the operator authors the experience instead of simply consuming it.
The promise is real, but so is the risk. Vibe coding without governance gives enterprises shadow IT at machine speed.
Speed was never the hard part of enterprise software. Trust was.
Gartner forecasts (paywall) that by 2028, prompt-to-app approaches adopted by citizen developers will increase software defects by 2,500%, and that 60% of vibe-coded applications will be retired for failing to deliver predictable business value. These are forecasts, not measurements, but they describe a familiar pattern of artifacts nobody owns or can audit, agents that fragment the data layer, and token spend nobody sees until the invoice arrives.
Human judgment persists, too. In Anthropic’s survey of 132 of its own engineers, AI touched roughly 59% of their work, yet more than half could fully delegate only 0% to 20% of it.
Most enterprises will respond with AI-assisted DevOps, which speeds up workflows that already exist while teams assemble context, permissions and governance by hand. Vibe coding sits at the other end of the agency-control curve: maximum agency, minimal control.
That creates a third requirement: Preserve the speed of vibe coding without accepting minimal control. The answer is governed execution, or controls that let teams determine what generated artifacts can access, change and deploy.
VentureBeat’s Pulse research offers a sobering baseline. Among 101 enterprises surveyed in July 2026, 68% had traced a confident but wrong agent answer to missing or inconsistent business context within six months, and 37% had seen it more than once. In a separate survey of 108 enterprises, 49% had, in the past year, shipped an agent that passed internal evaluations and then failed a customer. Both samples are small and self-selected, so treat them as directional.
The twist is more instructive. Organizations running or building a governed semantic layer, 63% of the sample, reported recurring failures more often (50%) than those without one (21%). VentureBeat reads that as detection; without shared definitions, failures still happen but get filed as model errors, user mistakes or nothing at all. A clean record often means nobody is looking.
The implication is that governance cannot be an afterthought to the demo.
The usual objection is that review gates slow the speed you are selling. I think that has the economics backward.
A vibe-coded app on a laptop is a prototype. Once it must read production data, write changes back and respect permissions, it needs connections, a permissions model and a map of where data lives. If those capabilities are provided centrally, teams don’t have to rebuild the same foundations for every artifact. Access control, auditability and cost metering can become inherited capabilities, while one person’s work can be shared, improved and reused.
DevOps governs the delivery pipeline. A comparable operating model is needed to govern intent, context and agent decisions.
An agent fed by a single tool sees a fragment, and fragments invite hallucination. Artifacts need a fabric that unifies observability, ITSM, network, cloud and security data without a rip-and-replace, plus standard connectivity such as MCP.
CMDBs and service maps go stale. Agents need a continuously updated model of dependencies and recent changes.
Rebuilding context mid-task burns tokens and compounds errors. Persisting intermediate results lets a human trace any decision to source data.
Generated artifacts should land where they are versioned, tested against historical incidents and approved under role-based permissions before they become enterprise capabilities.
Governance must reach cost, too. CFOs want business outcomes, not token counts, so spend should be tracked per agent, workflow and result, with smaller domain-specific models handling routine work where appropriate.
Most operational AI is advisory: It surfaces insight, then waits for a human. But AI-generated artifacts with context and tool access can turn software into an incident generator. It should therefore be earned in stages, with role-based guardrails, explainability and human approvals defining what an artifact may do alone.
The shift from UX to VX changes who the interface belongs to. Today, operational experience is consumption-based. An engineer opens a dashboard, hunts across consoles, correlates signals by hand and waits for a vendor road map. In VX, the operator states intent, such as “show me every service degraded by this change window and draft the rollback,” and gets a dashboard, workflow or agent grounded in live data that evolves with the environment. The gap between the person who understands the problem and the capability that solves could shrink from months to days.
Early results are encouraging, but company-reported. At Fabrix.ai, where I work, a Fortune 500 energy customer’s VPN root-cause analysis fell from hours to about three minutes, and our model estimates substantial annual savings across four use cases. Those are modeled estimates from one deployment, not an independent benchmark.
Nearly every platform shift in enterprise IT has followed the same pattern: ungoverned adoption first, painful correction second.
We can skip the correction this time. VX will arrive whether or not enterprises are ready. The leaders who benefit will treat governance as the enabler of speed, since people move fast only with tools they are allowed to trust.
Vibe coding gives operators the pen. Governance determines what they are allowed to write, and what can safely be run.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
