Picture your platform team's Monday standup. Three engineers have each built their own review helper. One lives in a script on a laptop, one is a prompt someone pastes into a chat window, and one runs only when its author remembers to start it. Nobody can say what any of them cost, which credentials they can reach, or who approved what. You are the person asked to make sense of it.
Kong has written up how it handled that same problem inside its own engineering organization, and the design choices are the useful part. In its post, How Kong Built an Agentic Engineering Platform, the company describes KSAI, Kong Skills for Agentic Intelligence. It "brings AI-assisted code review and implementation into GitHub, where our engineering work is actually done." The reason it exists will sound familiar to you: "Previously everyone was building something for their own, using their skills/prompts and tricks."
Here is what changed. The agents run in CI, "so closing a laptop doesn't stop the work." All model traffic goes through Kong's own AI gateway, for "observability governance" and to "hide LLM providers behind a single abstraction." That means one place to see calls and one place to swap a provider, rather than a pile of individual setups.
The post also reports one figure, from an A/B test it describes carefully. On 10 percent of runs, Claude Code and OpenCode reviewed the same change with the same model and effort. In the post's words, "we didn't see much of a difference in the cost of Opus runs, but surprisingly Sonnet runs were 40% cheaper on OpenCode." That is Kong's own measurement of its own workload, so treat it as one data point rather than a rule.
The isolation is the part you would copy. OpenCode runs in a bubblewrap sandbox with a read-only root filesystem, an isolated temp directory and separate namespaces. GitHub tokens, OIDC variables and runner credentials are "removed from the agent's environment." The model token is brokered outside the sandbox through a read-only file. And "Authorisation and approval remain separate boundaries": a control plane checks who may start a paid or writing run, and the pull request author and co-authors cannot approve their own changes.
Kong is also frank about its first attempt. Each subagent explored the codebase separately, and the post says: "In effect we were burning tokens multiple times for the same thing." If you have ever watched several agents re-read the same repository, you know the feeling.
Now the turn. APIs.io keeps a record on Kong, and this story is partly about what that record says. The Kong provider page lists 139 API pages. KSAI's model traffic runs through the Kong AI Gateway, whose per-request cost attribution the series covered last week, and agent traffic is governed by the Kong Agent Gateway. The agentic access profile maps 1,242 operations, 782 of them acting and 122 human-in-the-loop. The Kin Score is 71.8, in the exemplar band, and Agent Readiness is 46.9, agent-ready.
But the catalog also holds findings against Kong. The MCP server dimension is still unlit, a regression filed on the catalog's side earlier this month. Delegated identity is lit on Kong's record, while consent identity and dry-run mode are not. So the company that wrote up a careful internal design still has gaps in the public measurements.
The security design in the post reads like a list of the dimensions this series measures, applied internally: separate identity from approval, keep credentials out of the agent's reach, and attribute every model call. Kong has built for its own engineers the controls it sells to customers, and it said which first attempt wasted tokens.
What you take away for Monday: pull the agents out of laptops and into CI, route their model calls through one gateway, keep credentials outside the sandbox, never let an author approve their own change, and share the early mistakes so the next team does not repeat them. Then check your own record the way the catalog checked Kong's, because the gaps will show there too.
Read the original: https://apis.io/2026/10/26/kong-runs-its-review-agents-in-ci-behind-its-own-ai-gateway/