I’m currently working on a small team within Honeycomb where we’re building an ambitious new feature. We’re excited—heck, the whole company is—and even our customers are knocking on our door. The energy is there. With all this excitement, I’ve been thinking about a risk that—if I'm not careful—could severely hinder my team's ability to ship on time, celebrate success, and continue work after launch: burnout.

In a keynote at AI.Dev, Robert Nishihara (CEO, Anyscale) described the shift: A year ago, the people working with ML models were ML experts. Now, they’re developers. A year ago, the process was to experiment with building a model, then put a product on top of it. Now, it’s ship a product, find the market fit, then create customized models. The general-purpose generative AI models available to all of us today (such as ChatGPT) change the way work is done.

Today marks an exciting milestone at Honeycomb, and we're thrilled to share it with you. We officially launched our integration with Microsoft Teams, a step forward in our continuous effort to streamline and enhance your observability experience. Teams now joins our growing list of over 100 Honeycomb integrations.

Cloud-native developers and practitioners gathered from around the world to learn, collaborate, and network at KubeCon/CloudNativeCon North America 2023 between November 6th and 9th at McCormick Place in Chicago, IL—myself included. This wasn’t my first time attending—I’ve been coming to KubeCon since 2016—but it was easily one of the most exciting experiences I’ve had as part of the Cloud Native community.

ShipHero needed a robust, cost efficient observability platform to support DevOps, customer support, and more. Committed to timely service, ShipHero recognizes that the seamless performance of its software is paramount to customer satisfaction. To maintain this high standard, the development team needs the right data at their fingertips to quickly find and solve problems as they occur.

Now that we’ve had time to decompress from KubeCon, we wanted to do a writeup about our collective experience. Six of us spoke at the conference and Charity participated in a panel, so we included short talk recaps.

So you’ve taken a look at the core web vitals for your site and… it’s not looking good. You’re overwhelmed, and you don’t know what change to make because everything seems like too big of a project to make a real difference. There are so many measurements to keep track of and the standards cited seem even scarier. This is extremely normal. Web performance standards can feel impossible to meet for a lot of us.

Observability is important to understand what’s happening in production. But carving out the time to add instrumentation to a codebase is daunting, and often treated as a separate task to writing features. This means that we end up instrumenting for observability long after a feature has shipped, usually when there’s a problem with it and we’ve lost all context. What if we instead treated observability similarly to how we treat tests?

Ever since we launched Query Assistant last June, we’ve learned a lot about working with—and improving—Large Language Models (LLMs) in production with Honeycomb. Today, we’re sharing those techniques so that you can use them to achieve better outputs from your own LLM applications. The techniques in this blog are a new Honeycomb use case. You can use them today. For free. With Honeycomb.

A lot of reasoning in content is predicated on the audience being in a modern, psychologically safe, agile sort of environment. It’s aspirational, so folks who aren’t in those environments may feel like the path there includes doing “the new thing” or using “the new tool.” If you write software and your employer hasn’t caught up to all the newest, best ways to work, I hope this pragmatic post helps you sleep better at night.