In February, we launched the Companion Nicely-Architected Framework (PWAF), transferring our companion steering from static PDFs into AI-ready steering. For the primary time, it spans all three of our core companion architectures: Constructed-On, Related, and Information Collaboration. In my PWAF launch session at PKO, I talked concerning the innovation window: the second when the tempo of change at Databricks and throughout the market makes an entire new class of merchandise potential. Each functionality we ship is a chance to construct a brand new, differentiated product, typically with a brand new income stream hooked up.
As you possibly can see from the bulletins we’re making this week for the Databricks platform and our companion program, this window solely retains getting wider. We constructed PWAF as AI-enabled steering utilizing AI tooling. Our objective is to maintain tempo with what Databricks ships, transferring on the velocity of the product and the velocity of the AI market. So, this is a refresher on what PWAF is, a take a look at what we have added since February, and a preview of the place we’re taking it subsequent.
Anchored in Structure
PWAF begins with its structure heart. It builds on the well-architected rules you already know (the cloud Nicely-Architected Frameworks and the Databricks Lakehouse Structure) and focuses them on the patterns our companions truly construct with: constructing your product on Databricks, connecting to your clients’ Databricks to run jobs on their behalf, or sharing your information merchandise by means of the Databricks market. As companions more and more construct information and AI functions, brokers, and AI-powered experiences on Databricks, PWAF gives a typical set of patterns and requirements that assist speed up improvement whereas aligning with platform greatest practices.
We constructed steering for all three of those companion architectures: Constructed-On, Related, and Information Collaboration. For Constructed-On companions, it is anchored by Firefly Analytics, our reference implementation. The entire structure heart is AI-ready by design, so you possibly can level your coding agent at it and begin constructing.
Past the patterns, the structure heart spells out the technical requirements each integration has to satisfy, set to the identical bar our companion engineering crew validates in opposition to. It additionally exhibits you easy methods to instrument your resolution so your adoption and DBU impression are measurable, offering the info that may transfer you up the companion tiers and develop your GTM advantages.
Brick by Brick: What’s New Since February
As Stephen put it, we have added numerous bricks to the wall.
- The Databricks AI Companion Dev Equipment. We have packaged 15+ AI-developed expertise your coding agent can use, protecting integration patterns, telemetry instrumentation, and even prepping in your companion validation name. Each ability is totally examined and ships with its personal check suite. As an alternative of studying a sample and re-implementing it by hand, you hand the ability to the coding agent of your selection and let it construct in opposition to vetted, well-architected requirements. Again in February, I demoed constructing a JDBC integration with PWAF; it took round 20 prompts of back-and-forth with a coding instrument, utilizing my very own Databricks experience. With the Dev Equipment, that very same integration got here collectively in a single shot, and companions are telling us they’re seeing the identical on their very own builds. You spend your time on what makes your product totally different, not on easy methods to set up a connector or implement a user-agent.
- New and expanded sample steering. Since February, we have revealed net-new protection for Clear Rooms, software-defined storage, and Market apps, and refreshed our steering and requirements on the capabilities transferring quickest: Genie, Lakebase, and MCP server onboarding. As Databricks launches new capabilities and new patterns and requirements emerge we’ll maintain including extra steering to PWAF.
- Firefly is now open supply. Firefly Analytics, the reference implementation we constructed for Constructed-On companions, is dwell as a Databricks Labs repo you possibly can clone as we speak. It ships working examples of the arduous components of constructing an app on Databricks: auth, IAM, and SSO/SPN flows; enterprise-grade safety and scale; embedded apps; and AI. Take it as a place to begin, prolong it, make it your individual, and if you happen to discover a sample we’re lacking, inform us and we’ll add it.
Companion Engineering within the AI Period
The best way companions and Databricks construct collectively has modified, and it’ll maintain evolving with the AI period. What used to go between our engineers and yours can now run agent to agent, releasing each groups for the complicated structure issues we are able to solely resolve collectively.
Our companion engineers are builders, and that is how we construct alongside you: not with a single ability, however a full suite. Abilities deal with the routine integration work, structure steering tackles the more durable design choices, and reference implementations like Firefly give your agent a working instance to level at. In need of sitting at your keyboard, it is the closest factor to having our crew construct it with you, and our purpose is to assist each companion transfer quicker.
That is the multiplier impact we’re after collectively. Enabling one companion to construct quicker is linear; enabling each companion to construct deep, differentiated merchandise is how we purpose to show that into exponential progress throughout our joint buyer base.
The Window is Open
The innovation window is open proper now, and it rewards companions who construct deep and differentiated: utilizing extra of our platform to construct one thing your opponents cannot match, and that you simply could not construct anyplace else.
It is all dwell as we speak. Level your coding agent on the structure heart, pull within the Dev Equipment, and clone Firefly. Inform us what you ship, or what we’re lacking, by means of the Companion Portal.
We’ll maintain transport new patterns, expertise, reference implementations, and demos. Bookmark the structure heart and level your instruments at it, as a result of that is an ever-evolving framework, one which strikes quick and is barely going to maneuver quicker. It is early days, however we’re enthusiastic about what we’re constructing, and that we get to do it with a companion community this robust. Let’s construct it collectively, brick by brick.
