Calling a model provider directly means the service holds the provider's API key and builds its own traffic controls. Fine for one service, one model. Less fine as you grow.
Set up Kong AI Gateway, step by step: bit.ly/4eETw5L
How we decide when a frontier model is worth it for code review: model score = 100 × (F1 − review cost / V), with V the dollar value of going from F1 0 to F1 1. We run it at V=$1, $5, and $20.
In our own test, Opus comes out on top. It costs 7x more than GLM 5.3.
But we default to GLM 5.3 anyway: similar quality, far lower cost. 👉 bit.ly/4yM83Us
Kong retweeted
Insomnia 13.4 is about seeing more. 👀
Git changes grouped by endpoint. Secrets redacted everywhere in the UI. License usage your admins can actually read. Details in our blog: konghq.com/blog/product-rele…
Provider billing sees one API key. It can’t tell you which service, team, or customer spent what. A gateway sees every request with a caller identity attached. Here’s how to set one up: bit.ly/4srFn1u
Kong Event Gateway 1.3.0 can authenticate to @awscloud MSK backends with IAM credentials, including assumed roles. It also adds a Request Rule Validator policy that checks Kafka requests against rules you define.
bit.ly/3U8bK8V
We ran the same model at the same effort on two agent runtimes, Claude Code and OpenCode, for 10% of our code reviews. Opus cost about the same. Sonnet runs were 40% cheaper on OpenCode. The runtime is a separate choice from the model.
120,000 tokens vs. 1,200. Same workflow.
The difference is code mode: the agent writes code against the API instead of loading hundreds of tool definitions into context.
Marco Palladino on how to get there, and how to keep it governed: bit.ly/4ijQm8v
Kong is a founding member of the @ServiceNow Control Tower Alliance, alongside several others that will help every org adopt AI safely and securely. Every AI control tower needs a runtime arm. Here’s the news: bit.ly/4iMgM48
Every service that calls a model directly holds your provider key. One leak exposes the whole account, and rotation means changing all of them. A gateway holds the key and gives each service its own credential you can revoke.
Setup guide: bit.ly/4ioAbXF
Enterprise traffic to open-weight models: under 0.2% in May. 5% now. @sonicaghi explains to Hard Coded why Fortune 500 companies adopt slowly, then fast, and why every Global 2000 company becomes a neolab: bit.ly/4sqkHqB
📣 Every API + AI Summit 2026 session is now on demand: bit.ly/4A2hy37
Dive in virtually, then find us at @AWS re:Invent, booth 1485 (Nov 30 – Dec 4, The Venetian). Live demos of governing APIs, LLM traffic, and agents in production.
Book time: bit.ly/4kq0Eqe
4.2B tokens in a month. 162 consumers. 43 models. One governed, OpenAI-compatible endpoint.
@wwt_inc runs its LLMaaS in the AI Proving Ground on Kong AI Gateway. Here’s the whole stack, layer by layer: wwt.com/blog/inside-llmaas-p…
Kong retweeted
Kong CEO @sonicaghi reveals token traffic through its gateways is growing 140x year over year, approaching 1 quadrillion tokens:
"The number of tokens is growing exponentially. For the subset that we see, we are running about 500 trillion tokens, going to a quadrillion soon, because it’s growing 140x year over year."
"The cost per token is going down, and inference will still be exploding. We’re gonna live in this enterprise world that is more multi-model, cheaper tokens, cheaper levels of intelligence."
"The next pieces are, how do we secure it? How do we govern it? How do we cost control it?"
@kong
Context Mesh is here!
It's like a universal adapter for your agents: compose APIs, data, and docs into MCP interfaces agents can call directly. Agents discover tools just in time and retrieve only what the task needs.
Dive in: bit.ly/4jAUmDZ
We benchmarked blocking recall, precision, and F1 on real PRs to determine when a frontier model is worth the price for code review.
Opus costs 7x more per review than GLM 5.3. We added a triage step that matches model and effort to each PR, and median review costs dropped from $3.93 to $2.47.
Here's the story: konghq.com/blog/engineering/…
Most enterprise AI projects don’t fail at the model. They fail at the infrastructure underneath it.
Find us at booth P13 at AGNTCon + MCPCon (San Jose, Oct 22–23) for an in-depth demo and a chat with a Kong exec.
A haunted house, but it’s every team building its own auth, routing, and fallbacks, then calling it governance.