Vibe coding is coming to the enterprise. UST rewrites or rejects 40% of the output

UST launched Codon on September 1st, an AI workspace for building and changing applications across ServiceNow, SAP, Salesforce, Oracle Fusion and Workday, and Chief Solutions Officer Kailash Attal told RuntimeWire that UST experts rewrite or reject roughly 40% of Codon's generated work, a rate he called "healthy for the product's current stage." A production pilot went live on September 9th and is moving toward a 50-user readiness gate, while UST says Codon cut the time from defined requirement to reviewed, deployable result by about 50% across more than 50 internal builds on more than 10 projects. Every Codon deployment passes through approval gates, audit trails and mandatory human review, positioning AI as a draft generator rather than an autonomous replacement for the practitioners accountable for enterprise systems.

Exclusive: Vibe coding is coming to the enterprise. UST rewrites or rejects 40% of the output UST Chief Solutions Officer Kailash Attal tells RuntimeWire how Codon brings plain-English building to five major enterprise platforms while keeping every deployment behind approval gates, audit trails and mandatory human review. By Ryan Merket https://runtimewire.com/author/ryan-merket · Published Why it matters Codon shows how IT services providers can productize AI without removing their consultants. Generated code becomes cheaper; governance, review and contractual accountability become the service being sold. UST is asking enterprises to adopt AI coding as a governed delivery process, not an autonomous replacement for the people already responsible for their systems of record. The services company launched Codon https://www.ust.com/en/alpha-ai/ust-codon on September 1st as an AI workspace for building and changing applications across ServiceNow, SAP, Salesforce, Oracle Fusion https://runtimewire.com/models/openrouter/fusion and Workday. A developer describes a requirement in plain English and identifies the target platform. Codon plans the work, routes it through the relevant Model Context Protocol server and platform software development kit, and assembles a proposed change. A senior UST practitioner then reviews the output before it enters the customer's approval and deployment process. Chief Solutions Officer Kailash Attal told RuntimeWire in written answers that UST experts rewrite or reject roughly 40% of Codon's generated work. He described that rate as "healthy for the product's current stage." That figure defines the enterprise adoption model more clearly than the launch language does. Codon generates the first pass, governance limits what it can do, and experienced practitioners remain responsible for deciding what is fit to deploy. Enterprise adoption remains at the pilot stage Attal said a production pilot "went live on September 9th" and was moving toward a "50-user readiness gate." UST has not identified the customer, leaving Codon's most concrete operating data inside the company's own delivery organization rather than in an independently documented enterprise deployment. UST says Codon reduced the time between a defined requirement and a reviewed, deployable result by about 50% against its historical benchmarks for similar ServiceNow and Salesforce engagements. Attal said the measurement covered "more than 50 internal builds across more than 10 projects." The comparison is narrower than the broad productivity figures that often accompany coding-agent launches. It covers a specific stage of delivery, uses UST's historical work as the baseline and includes expert intervention. UST's September 1st launch release https://www.prnewswire.com/news-releases/ust-codon-brings-governed-ai-development-to-five-leading-enterprise-platforms-302866063.html also cited an application completed in about a week, compared with an estimated three to four months under the traditional process. The 40% rewrite or rejection rate helps explain how UST can promise deployable outcomes from generative models whose initial output frequently needs work. Codon is less an autonomous enterprise developer than a productized delivery pipeline: AI supplies a draft, while UST practitioners retain the final quality obligation. Governance is the adoption layer Codon's public materials describe controls designed for companies that cannot let a general-purpose coding agent make untracked changes to payroll, permissions, customer records or financial workflows. UST says every tool call is classified as a read or write operation, with unknown or ambiguous actions failing closed. Writes are attributed to individual developers where the underlying platform supports that identity model. Production changes pass through an approved-build phase, while destructive actions require explicit intent. Codon also records activity in what UST describes as a tamper-evident, hash-chained audit trail. Customer environments run as single-tenant deployments on AWS Fargate with default-deny outbound access, encryption in transit and at rest, and a warm standby in another region. For enterprise adoption, those controls are as consequential as the generated code. The relevant standard is traceability and reversibility, not whether a model can produce a plausible result in a demonstration. Codon offers chat and terminal interfaces governed by the same policy and audit engine. The chat route is intended for workflows, catalogs, configuration and documentation. The terminal gives developers a per-user shell for platform command-line tools and software development kits. Human review is also part of the contract Codon previews expected AI consumption, implementation effort and total cost before a build begins. UST combines model token rates with estimates for expert review, then provides a quote with an upper limit. Who carries an overrun depends on the engagement. Attal said UST absorbs excess AI and platform costs under managed or outcome-based contracts. Customers pay consumption charges under time-and-materials arrangements, with UST providing an estimate before work starts. The estimator therefore functions as a services pricing mechanism as much as a software feature. UST owns Codon and its integrations, but the commercial model remains tied to UST's delivery staff. The volume and difficulty of generated work determine the human review burden, making the disclosed rewrite rate relevant to margins as well as deployment speed. That arrangement reflects how UST expects enterprises to adopt AI: through an existing vendor relationship, with review, audit and financial accountability packaged around the model. The customer is not simply buying access to an agent. It is buying an accountable delivery process in which the agent performs part of the work. Platform vendors want the same enterprise work UST is entering a market where the underlying software vendors are building their own AI development surfaces. Salesforce markets Agentforce Vibes https://www.salesforce.com/agentforce/developers as a natural-language system that can plan, code, test and deploy Salesforce applications. SAP's Joule Studio https://news.sap.com/2026/05/new-joule-studio-enterprise-scale-agentic-development/ generates requirements, technical specifications, code scaffolding and tests from business intent grounded in SAP data and processes. Those vendors control the metadata, security models and development tools inside their platforms. UST's counterposition is breadth: one workspace and governance layer spanning five large enterprise systems, backed by practitioners who already implement them. That cross-platform model is most relevant when a business process crosses several systems. A hiring workflow, for example, may begin in Workday, trigger access provisioning through ServiceNow and update financial or sales records elsewhere. UST is betting that customers will prefer a shared control plane and one accountable delivery partner over a collection of vendor-specific AI builders. Codon's early figures show both the appeal and the constraint. UST says AI can halve a defined portion of its delivery cycle, but human experts still replace or discard about two-fifths of the generated work. For now, enterprise AI adoption means keeping that gap inside a governed, billable process rather than pretending it has disappeared.