Build at agent speed. Safeguard production.
AI code review in production at FPT Cloud, and the agentic operating model it opens up.
Precision
~97%
Precision observed in the evaluation. Engineers acted on what the reviewer reported rather than triaging it. In this evaluation false positives were rare, so no new review queue appeared.
Backend logic coverage
Full
Backend logic coverage in the evaluated set. Defects observable in the code itself were detected, security findings among them, before human review or QC testing.
About the customer
About FPT Cloud
FPT Cloud is a comprehensive cloud computing platform with nearly 100 services built and operated directly by Vietnamese engineers. The platform serves the digital infrastructure needs of enterprises in Vietnam and abroad, from Infrastructure as a Service (IaaS) and Platform as a Service (PaaS) to security and advanced AI development services. Running on advanced GPU infrastructure and certified against national and international security standards, FPT Cloud helps enterprises shorten deployment time, optimise operating costs and stay ready for the specific requirements of every industry.
The ambition
Building the world-class Enterprise Agentic Platform together, to safeguard production in the AI era
Code review is the first gate. The same governed model extends across the production estate: agents that investigate, decide and act under human control, on the platform thousands of enterprises run their workloads on.
What it changes
Four shifts in how software is built and defended
AI-assisted development raises throughput. A gate on every merge request means the business can take that speed without absorbing unmeasured risk.
On a shared platform, one cross-tenant defect is a compliance event, not a bug. Catching that class at the merge request keeps it out of the release entirely.
Standards that lived in senior engineers' heads become a curated rule set applied to every change. Review quality stops depending on who is available.
Findings are verified and the process is measurable, so quality can be reported to customers, auditors and the board as evidence rather than intent.
The operating model
Every layer does what it is best at, and nothing twice
Review (AI code review)
68.6%
QC
25.5%
AI review is the first gate, at the least expensive point in the pipeline. Human review and QC keep what only appears when the software runs.
Adoption
Value from day one, pipeline optimised in the first month
Day one
Running on live merge requests
No pipeline rebuild and no waiting for a training phase. The reviewer sat in the existing workflow and returned findings on real changes from the first day.
First month
Tuned to how the team works
Requirements, issue context and the team's own engineering standards were loaded into the review. Accuracy climbed as context arrived, and the pipeline settled into its production shape.
Next chapter together
From governing what gets built to operating what runs
Co-build the core technology of Agentic to help thousands of FPT Cloud customers operate the production environment.
Specialist agents get the context and skills to investigate, decide and act across the cloud, Kubernetes and data platforms you operate, with approvals, guardrails and a complete audit trail on every operation.
Every team, one entry point
SREs, platform, application, support, IT ops.
One shared memory
What one agent resolves, the rest can use next time.
Faster incident response
Saving the operation effort
Agents own production, under your policy.
Focus on what matters most
Talk to the team
Put the same quality gate on your merge requests, then extend the governed model into the environments you run.