We’re excited to launch Canyon, now in closed beta. 

Canyon is a self-service AI for developers. It’s reliable, stays within secure guardrails and boosts productivity. It clears blockers, accelerates developers' velocity and dramatically reduces cognitive load. 

Canyon is designed for zero hallucination in execution. It combines the power of large language models (LLMs) with a comprehensive graph of your entire estate and a deterministic backend that assumes the users’ permission set. Admins, Platform Engineers or Operations teams can configure “golden paths” as code. As developers request things (access to parts of the estate, resources, apps, environments, monitoring information, etc.) Canyon will identify the correct path, keep the human in the loop and execute it on behalf of the user. This approach makes Canyon both reliable and secure, allowing large enterprises to adopt it at scale. 

While there’s virtually no limit to what paths you can enable with Canyon, here are a few examples that users loved during early tests:

• I’ve just joined X team on Canyon. Can you give me a rundown of how the apps work? 
• Can you create an ephemeral environment for this PR?
• I have to store videos and photos from users. Can I do that in my app?
• There’s a bug in the DNS, it’s routing to a weird place 
• I need permission to see XYZ app
• Are there any differences between production and development? Show them as a table.

And for more infra/ops focused users: 

• What paths are available in this application?
• Are there any differences between production and staging?
• Show me every application using ${some-resource} as a graph
• Show me all the S3 resources and which environments they’re used in as a table.

We truly believe that, when applied correctly, even today’s AI has the power to transform the Software Development Lifecycle fundamentally. As long as we recognize that today’s AI technology is a.) a high-context and user-centred UX layer, and b.) excels when it’s trained on large datasets and asked to make a choice from a finite set of options, then we can really start to design systems that play to its strengths. 

While Canyon is still in its infancy, it’s the culmination of a decade-long journey: one rooted in serving developers, helping them move faster and more autonomously, while maintaining secure guardrails.

Why Canyon

The debate about the optimal design and distribution of responsibilities in engineering organizations has been raging for decades. Ever since the dawn of the “you build it, you run it” movement, the industry has fluctuated between specialization and generalization. United by the shared dislike of the operations-centred model of the 2000s, we’ve worked towards calibrating how to balance the need for autonomy, speed, cognitive load of building and managing globally scaling infrastructure. 

The idea behind Platform Engineering was born to resolve this fundamental tension. Early 2020 we launched internaldeveloperplatform.org and organized Meetup groups in 25 cities around the world to promote this movement. By 2021, we published platformengineering.org, and in 2022 we introduced PlatformCon. The development of Platform Engineering has been nothing short of a miracle to us. 

Throughout the years, we’ve always promoted the approach of paved roads and golden paths to streamline the recurring activities of engineers. Yet we continued to struggle with what the optimal UX would be to consume these paths, as well as how to put it on top of an Internal Developer Platform (IDP). We launched a portal, a CLI and a code-based interface, but have found that those interfaces were either conceptually difficult to understand for the user or didn’t optimally blend into the current workflow. 

We believe that conversational interfaces and AI can optimally bridge the gap and help the user pick, and walk, the right path at the right time.

How does it work?

We’re shipping the first version as a plugin in your IDE to ensure we’re meeting the user where she is; more interfaces including Teams and Slack will follow. Execution of a request happens across 6 steps: 

1. The user articulates in plain natural language what information she needs, what to update or what to create from scratch. 
2. Canyon enriches this with additional context and forwards this to any LLM (can be hosted by the Enterprise or via a hosted service. 
3. The model can pick from a set of paths that are provided as code in a repository using the Model Context Protocol. 
4. Canyon then proposes this action to the user and asks to accept the choice. 
5. Canyon checks the user's permission set (what rights does the user have across projects, apps, environments etc.). 
6. Canyon executes the path and informs the user. The output of one path might now be the input of another. 

In this early version we are shipping only a couple of default paths. We differentiate between query paths (consolidate information across a scattered estate) and edit paths (update and create new things). Additionally, you can of course, a.) add your own, b.) we’re setting up a Discord server for you to propose and vote for new paths, and c.) we’ll continue adding paths as we go. 

Your path into the Canyon

We’re thrilled to release this first version of Canyon in closed beta. If you’re interested in setting it up and providing feedback, reach out to us. For the next few weeks we’re really focusing on learning and iterating through your feedback, simplifying the backend and making it as little intrusive, and as easy to set up, as possible. 

We cannot wait to see all the great paths you’ll pave through the Canyon!