Back from #KubeCon in Chicago. My notes below: - Valuable debate about what "open source AI" really means: Is it the model, the training data, the weights, the feedback, or all of the above? If you only open source the model itself, is it really "open source"? Good to hear from Jim Zemlin that the Linux Foundation is working on this. As my colleague Manjunath Bhat has said, an LLM is really only "open source" if you can recreate it entirely. For that, you need the data, weights, and feedback. - Software supply chain security is big. The CNCF has conducted audits on Argo and Prometheus. Good to hear about the work of the Open Source Security Foundation at KubeCon. - Many of the vendors we cover in our Market Guide for Service Mesh were there, and I enjoyed the "service mesh battle scars" session moderated by Keith Mattix II from Microsoft. There is clearly a disagreement between vendors about whether to use mTLS for both handshaking and transport encryption, or just for handshaking. The debate between Isovalent and Buoyant about the appropriate usage of mTLS was particularly spirited. Similarly, there was a good debate on eBPF, particularly whether it is useful for telemetry only, or also could be used for traffic management, with solo.io weighing in. Finally, the "sidecar vs sidecarless" also drove a wedge between the vendors too. Service mesh is still an evolving area. - BackstageCon was my focus on Monday, particularly since we published our Market Guide for Internal Developer Portals, including Backstage. The day kicked off with Roadie talking about Backstage deployments, then other vendors including Red Hat and VMware speaking about their Backstage plans. My overall takeaway is that we're at nearing peak Backstage hype now, but nowhere near peak Backstage deployment. Backstage is still difficult to deploy, with one speaker mentioned 70 individual steps to get it up and running. It's also difficult to drive team-wide adoption (this goes for any developer portal, though). I had some good conversations with vendors like Port ⛴️ , Harness, and Cortex who will be interesting to watch. - And something I *didn't* see: much debate/controversy/blowback about recent license changes.
Zohar Einy
10mo
Great recap. The way I see it, the key to adoption is making sure the Portal fits into your organizational DNA, the reason Backstage got adopted at Spotify is since it was built FOR Spotify. A key consideration for implementing a portal is to make sure you remain the Product Manager and you have enough freedom to compose an IDP that fits exactly to your engineering teams. At Port ⛴️ , we invented the Blueprints, relations and UI toolbox to build YOUR portal. demo.getport.io
Anish Dhar
10mo
Thanks for coming by Mark and agree on your comments around Backstage. IDP’s like Backstage give you a lot of freedom to compose something that should fit your engineering team but the reality is you’re stuck configuring YAMLs and relations all day every time you want to add a new component. After 6 months you’ve overengineered a platform and also not seen the ROI you want to see to justify the internal costs of building. With Cortex there’s a ton of automation built into the onboarding process and teams still get tons of flexibility to build on top of the platform with our plugins framework. It’s how customers like SoFi, Grammarly, and more got started and found value in days. https://tour.cortex.io/share/6syjq8zytfqy
Himanshu Mishra
10mo
Good take on the Backstage deployment and adoption hype. It has only started. IMO it is almost an art to have a well-adopted successful IDP - and is more of a cultural problem than a technical challenge. As Harness, we've already made it easy to get started with Backstage/IDP - more to come on adoption strategies and how we're making our customers successful.
James Croft
10mo
The first point on here is such an interesting topic! While I somewhat agree with your colleagues thoughts, what value does someone get from being able to recreating a model? Unless of course the intention is to tweak it in someway perhaps.
Jason DeTiberus
10mo
While I think it's important for LF to help move AI projects they are associated with to a more comprehensive model of open, I'm more excited by the work being done by the Open Source Initiative in this space: https://blog.opensource.org/closing-the-2023-rounds-of-deep-dive-ai-with-first-draft-piece-of-the-definition-of-open-source-ai/
Pedro Quiñonez
10mo
Total agree at least one year algo we tried to deploy the Backstage and was to dificult, very interesting hear that after one year the challenges are similar. If the community expect the project adoption its necessary enhance the backstage deployment.
Tricia Cooper
10mo
I am glad you were able to make it Mark. If you need anything from the CNCF analyst relations team in terms of follow up please don't hesitate to reach out.
Oliver Cronk
10mo
Colin Eberhardt you'll probably find this interesting
Mohamed Tleilia
10mo
Amir SBOUI Hamza Bensalem Zied FAKHFAKH
CEO & Co-Founder, OpsLevel
10moSpot-on regarding Backstage's adoption difficulties. Did you swing by OpsLevel's booth at Backstagecon, Mark? Our Developer Portal has been built with a focus on quick time-to-value. Customers of ours - like Duolingo & Hootsuite - automatically build out their IDP catalog through integrations and automation, and avoid painful YAML like all the other players in the space seem to end up requiring. (I'd love to show you what we've built!)