Author profile
Khai Trinh
Engineering, CloudThinker
Khai Trinh leads R&D at CloudThinker, focused on improving the accuracy and reliability of the agentic system.
He writes about chat and dev tool integrations (Microsoft Teams, Slack, GitLab, Azure DevOps) and the Custom Agent builder that puts agentic workflows into the tools teams already use.
Posts by Khai (6)
Product
Your Zabbix Sees the Problem. Who Closes It?
Zabbix is great at noticing. Closing the problem is still a person clicking through the frontend. Why flapping triggers, stale dependency maps and fixed actions keep the queue growing, and how CloudThinker reads Zabbix over its API to triage, correlate with recent changes, act with approval and verify each problem stays closed.
Product
Meet Your Custom Agent
Generic AI doesn't know your stack, your playbook or your team. A Custom Agent does: name it, teach it, plug in only the tools you allow, and your whole team @mentions the same AI teammate in Slack, with every write going through Auto Mode.
Product
Investigation Anywhere: CloudThinker Now on Microsoft Teams
Most Teams channels carry the alert while the investigation happens somewhere else. CloudThinker on Microsoft Teams investigates in the thread, routes notifications per channel, and links to the web app whenever a change needs approval.
Product
Introducing CloudThinker Review for Azure DevOps
Branch policies count approvals and green builds, but they can't read code. CloudThinker Review now reviews Azure DevOps pull requests: a scoped PAT, one service hook per trigger, and findings ranked by severity posted as PR threads.
Product
We Tried Every AI Sandbox. Then We Built Our Own.
Hosted sandboxes couldn't reach our private APIs, self-hosted VMs couldn't pause, and the best open-source option lacked a production layer. We built persistence, tiered pause/resume, network filtering and hardening on Alibaba's OpenSandbox, and open-sourced it as OpenSandbox on EKS.
Product
AI Code Review Benchmark: 37 Real Bugs, Six Tools
We ran six AI code reviewers on 37 real production bugs from Cal.com, Sentry, Grafana and Keycloak, following Greptile's published method. CloudThinker caught 30 and Greptile 29. The method, the line-plus-impact grading bar, and why a self-run benchmark should be read in bands.