Systems | Development | Analytics | API | Testing

How We Used Agentic AI to Fix Kong Gateway's Flakiest Tests

Each change to Kong Gateway's codebase triggers a comprehensive test suite that runs more than 17,000 * 2 = 34,000 test cases among the two primary architectures (x86 and ARM) we support. This process takes about 23.5 hours on a single machine. But we don't wait that long. A large fleet of machines runs the suite in parallel, and we shard the work aggressively so each commit finishes in a fraction of that time. That setup works well, right up until flaky tests get involved.

Kong and Persistent: Take the Risk Out of Migration

Kong and Persistent Systems partner to make migrating off old API management platforms faster and lower risk Legacy API management platforms were built for a different era. They weren't designed for microservices, multi-cloud deployments, or AI workloads. They're expensive, rigid, and hold engineering teams back. The problem is that migration has always felt hard. APIs are load-bearing infrastructure. Policies are complex. Risk is real. So the old platform stays, and the technical debt compounds.

API Gateway vs AI Gateway - What Actually Changed?

Kong's AI Gateway applies the same architectural pattern as the API Gateway — now governing LLM, MCP, and agent traffic at the infrastructure layer. Just as API gateways abstracted rate limiting, auth, and caching across microservices, AI gateways do the same for large language models and agents — with token budgets, semantic caching, and semantic routing replacing their REST equivalents. Kong breaks this into three layers: LLM Gateway, MCP Gateway for tool calls, and Agents Gateway for agent-to-agent traffic.#Shorts.

Tokens Per Watt Is the Real Limit on AI Revenue

Most AI revenue will flow through tokens — and the two bottlenecks are tokens per watt (energy cost) and tokens per second (throughput). Tokens per watt determines how much output you can generate from a fixed energy supply — already constrained and getting tighter. Tokens per second sets the ceiling on how fast that revenue can flow. Kong's AI Gateway optimizes both at the connectivity layer: semantic caching and semantic routing increase token output without adding watts or latency.#Shorts.

Are Microservices Dying?

LLMs are absorbing the business logic of microservices for agentic use cases — but both patterns will coexist in enterprise infrastructure for a long time. Cloud-native infrastructure (microservices + APIs) keeps powering web and mobile experiences. The agentic layer — LLMs, MCP tool calls, and context traffic — runs in parallel, activating the same APIs and CRUD operations underneath. Kong manages both swim lanes: the API traffic between clients and microservices, and the context traffic flowing between agents and LLMs.#Shorts.

CLI vs MCP: One Gives Speed, the Other Governance

CLI offers speed and developer freedom for API access; MCP provides centralized security, governance, and observability at enterprise scale. With CLI, credentials live on the developer's local machine and audit trails are shell-only — fast, but ungoverned. MCP adds authentication, centralized policy enforcement, and observability across all API calls, at the cost of some speed and higher token consumption. Kong's MCP Gateway is built for teams that need the governance trade-off without giving up too much velocity.#Shorts.

Anthropic Acquires Stainless. What's It Mean for AI Connectivity?

Every few months, a frontier AI lab makes a move that says the quiet part out loud: agents are only as useful as the systems they can reach. The latest example is Anthropic's acquisition of Stainless, the company behind the tooling that turns API specs into SDKs and MCP servers. Anthropic's own framing is direct. Agents need to connect to data and tools, and the path from an API to an agent-ready interface needs to get shorter. We agree. We've been making a version of this argument for two years.

How Headless Software Powers the Machine Internet

Software is going headless: the internet is shifting from GUIs built for humans to APIs, MCP servers, and CLIs built for machines and agents. Machines will consume the internet at a scale 1,000x greater than humans — more agents will exist than people, and programmatic access moves far more data than any click ever could. This transition requires API and AI infrastructure capable of moving terabytes at a scale never built before. Kong provides the connectivity layer for this machine internet — the infrastructure between agents, LLMs, and the services they consume.#Shorts.