WunderGraph

Monday.com rebuilds their API edge for the agentic era with WunderGraph. → 10B+ requests/month → 2.5x the traffic → p99 edge latency: ~490ms → 44ms → 92% less CPU → 80% less memory "We believe agents will become the main consumers of the monday.com API, so we’re optimizing the edge for agentic flows: many small, fast calls made by software on behalf of a person." - Michał Budziak It's been a pleasure working with Shanee Radzewsky, Michał Budziak, Adam Rutkowski https://lnkd.in/eNvYXHkJ

monday.com now runs more than 2 billion API requests a week through a new API edge with WunderGraph Cosmo as its router. Their engineering team rebuilt the API edge because growing AI agent traffic was pushing it to its limits, requiring repeated increases in CPU and memory resources. They split the edge into a gateway for authentication and other request checks, and WunderGraph Cosmo for query routing. Compared with a year ago, the edge handles 2.5× the traffic with 92% less CPU and 80% less memory. The time the edge adds to each request at p99 is also 11× lower, from about 490 ms to 44 ms. Thank you to Michał Budziak and the monday.com team for the collaboration and for sharing the engineering behind the rebuild. Read their full write-up: https://lnkd.in/gaaMrijT

LLMs make people collaborate less in big enterprises! (It's really bad, but not what you think) Not my opinion, it's an observation from someone working in big tech. So what's going on? It's very simple. In the past they had to collaborate and ask another team to add a capability to one of their services. Now you just ask an LLM to generate a solution quickly. Read through docs and foreign code, no problem. The issue? It's not primarily a problem that people collaborate less, although it's sad, but the real problem is API sprawl. You lose complete control over your architecture and service landscape. A possible solution? GraphQL Federation with a strong governance layer (https://lnkd.in/exNWAJGD)

At WunderGraph we believe that GraphQL Subscriptions have a GAP! Or more specifically, we're creating one, lol. No, it's not what you think. I mean that we're going to make a GraphQL Auxiliary Proposal, a proposal that just doesn't really fit into the spec. So what's wrong with Subscriptions? A few years ago I came up with the idea of EDFS: Event Driven Federated Subscriptions. GraphQL Subscriptions powered by Kafka or other event brokers/streaming services. EDFS didn't seem to be a good name though, some people made jokes about this ED file system. So we renamed it to Cosmo Streams, alright. But what's the problem? It's about delivery guarantees. Streaming services like Kafka use cursors and can give you at-least-once delivery guarantees. However, when combined with GraphQL subscriptions, we're losing all delivery guarantees. A client has no way to express if a message was successfully delivered or not. If the client reconnects, where do we start delivery? The spec says nothing about delivery. It's a problem for our customers. If a user is making a bet on a race, or if they are buying stock, or they want to bid on an item, or get the latest state on the stock of a product, in all of these scenarios it's important that the user can trust that they have the most up to date information. Have you experienced this GAP as well? Did you know about GAPs yet? (Check the comments for a link to the repo)

5x growth YOY in usage with a non AI native solution, how's that possible? Cosmo is an open source GraphQL Federation solution. Router, Schema Registry, and so much more, the full suite. What problem does this solve and why so much growth? If you're a large org and you have many API consumers and even more backends, your API landscape can be an absolute mess. With Cosmo, it feels like a monolith for the API consumers, but one that can easily adjust to different use cases. On the backend side, teams get autonomy without losing governance. Every team can work at their own pace and stack, but we're preventing API sprawl by design. If you want to execute your AI strategy successfully, you have to get your API sprawl under control. Enterprises know this, which is why we're growing so fast. In addition, that unified "virtual monolith" on top of your micro services makes agentic workflows much easier. A unified graph is the perfect foundation for MCP with centralized governance and security. What excites me the most is that we're now entering the next stage of this pattern: Agentic Federation. Very soon, Federation will enable agents to speed up development dramatically. From schema design to reviews and governance. Federation is no longer about GraphQL at scale, it's about allowing agents to work autonomously on complex domains with centralized governance. And the best part, that's no longer just a vision. The market is pulling. Demand is here. Enterprises are ready.

The launch date is fixed. The monolith is not. Wrap existing services and federate the new domain without a rewrite.