vi
en
ja
FPT AI Mentor is a multi-tenant SaaS platform that helps enterprises train sales, customer-service, and collections teams through AI-simulated conversations. Since early 2026, its product owners, business analysts, developers, and testers have used Claude Desktop, Claude Code/Cowork, Claude Opus, and Claude Sonnet to support requirements, technical design, coding, testing, and defect resolution. The platform has been live in production since 7 July 2026.
Business Challenge
The team needed to increase delivery speed without lowering quality. In the previous workflow, creating prototypes and product requirements, then translating them into code and tests, required substantial manual effort and cross-functional handoffs. The team therefore needed a repeatable approach that could accelerate early-stage product definition, strengthen alignment between requirements and validation, and fit within its existing continuous-delivery process.
Claude Embedded Across the Delivery Lifecycle
The team adopted Claude Desktop as a common AI workspace, using Claude Opus and Claude Sonnet according to task complexity. Product owners and business analysts use Claude to turn approved prototypes into product requirement documents. Developers use it for architecture exploration, technical design, coding, unit-test generation, and bug fixing, while testers use it to create code-based end-to-end tests for critical user journeys. The team is also extending Claude into activities that connect development with deployment and production support.
Claude usage is formalised through an AI-executed Definition of Done: each feature must include an AI-written PRD, AI-written unit tests, and AI-written critical-path end-to-end tests. This turns AI assistance into a consistent delivery standard and keeps requirements, implementation, and validation aligned to the same feature scope.
Phased Implementation
Implementation follows a three-phase roadmap. Phase 1, from January to 30 September 2026, assigned one repeatable task to each role: PRD creation for product owners and business analysts, unit-test generation for developers, and critical-path end-to-end test creation for quality assurance. A weekly “Do > Share > Improve” loop helped the team apply Claude to live work, share effective practices, and refine its workflows before expanding automation.
Phase 2, scheduled for 1 October to 31 December 2026, expands Claude to more complex work and targets an integrated Architecture > Code > Test > Fix > Deploy workflow across Product, Development, Testing, and Operations. Phase 3, planned for the first quarter of 2027, extends the model to research, monitoring, hotfix and rollback support, and documentation synchronisation.
Operating Model and Governance
The FPT AI Mentor product team owns and delivers the initiative, with Claude used across the full delivery organisation rather than by a separate specialist group. Product and engineering owners remain accountable for reviewing AI-generated outputs and for production decisions, supporting consistent practices without transferring responsibility to the tool.
Outcomes and Business Value
Claude is now used by 100% of the product team across the software lifecycle. Since the platform entered production on 7 July 2026, every feature has been delivered under the AI-executed Definition of Done, with an AI-written PRD, unit tests, and critical-path end-to-end tests.
The measurable gains are consistent across two levels of delivery. Prototype creation time fell from two working days to four hours. Across the broader prototype-and-requirements phase, cycle time decreased by approximately 40–50%. At the end-to-end level, the release cycle was reduced from two weeks to one week. Together, these results indicate faster product definition and a shorter path from feature conception to production release.
Key Takeaways
The FPT AI Mentor case demonstrates that enterprise adoption of Claude becomes more effective when AI is embedded in explicit delivery standards rather than used for isolated tasks. The approach combined role-specific use cases, a shared improvement loop, and accountable human review. This foundation has already shortened prototype and release cycles and provides a practical path toward broader lifecycle automation.
