Conference summary
KubeCon EU 2024 Paris: Key Takeaways
AI, Platform Engineering, and Product Thinking…
It’s time to once again reflect on my travels to the recent KubeCon EU event, which was held in Paris, France. As usual, I learned a lot from the keynotes and sessions — with a strong focus on platform engineering — and met many super interesting people from the community. With over 12,000 people attending, not to mention 100s of sponsors, it was busier than ever.
Here are my key takeaways for KubeCon EU 2024:
- I, for one, welcome our new (cloud-based) AI overlords
- A call for responsible innovation i.e. “don’t forget about cost and sustainability”
- Platform engineering takes center stage
- Product thinking FTW!
- More interest in developer experience and inner and outer dev loops
- Security continues to be big business
- End-user stories are moving up the stack
- The tool and framework bundling continues
- Wasm: A hot topic, but with some uncertainty
- Dapr is fast becoming the cloud native ESB (in a good way)
I’ll explore each of these takeaways in detail, and I’ve also included a bonus “GTM takeaways for KubeCon EU” section at the end of the article. As the French say, allons-y!
I, for one, welcome our new (cloud-based) AI overlords
Many of us were surprised by the lack of AI/LMM content at KubeCon NA Chicago in December last year. Well, KubeCon EU Paris made up for this… And then some! Practically all of the opening day keynotes were AI-focused (to some chagrin!) There were also dedicated AI labs running throughout the week and many breakout sessions focused on AI/LLMs from all angles. And I lost count of the number of sponsor booths where I saw the words “AI”, “LLM”, or “Generative” mashed up with cloud native buzzwords.
Jokes aside, the main message I took away was that the CNCF ecosystem is very much open for business regarding AI. The end-user talks I saw throughout the event mainly focused on using cloud native tech for inference (rather than model training); particularly inference at or near the edge. This echoed a previous pithy statement by Clayton Coleman: “If inference is the new Web app, then Kubernetes is the new Web server’.” There were also plenty of related “picks and shovels” for the K8s-based AI builders being demonstrated both on the keynote stage and sponsor hall.
It was super interesting to see NVIDIA take to the stage as the second keynote of the opening day with “Accelerating AI Workloads with GPUs in Kubernetes”. They took care to weave in the Kubernetes story, but I couldn’t help but think this was primarily a hardware/infrastructure play, which may have been a bit lost on this audience. It was also a solid reminder that NVIDIA currently runs the show when it comes to GPUs and AI chips.
On a related topic, I was listening to the All-in podcast over the weekend, and the besties wondered aloud where innovation and value capture would occur within the four layers of the AI stack: infrastructure, (foundational) models, dev tooling, and applications. If NVIDIA has captured the infrastructure layer, our next keynote highlighted another company working on capturing the model and dev tooling layers: Microsoft.
Kicking off a topic we’ll cover later, Microsoft was the first to announce an AI-themed “bundle” at the event, with the “Kubernetes AI Toolchain Operator (Kaito)”. Kaito aims to automate the AI/ML inference model deployment in a Kubernetes cluster, and targets models such as falcon and llama2.
Continuing on this theme, Ollama also featured heavily throughout the show and appeared to be the de facto way to run large language models locally. As we were in Paris, the Mistral AI folks got their props, too. Even if you couldn’t make it to KubeCon, there were several AI-themed meetups running nearby:
There’s no denying the pace of innovation is happening at breakneck speed in the domain of AI. Many end users I chatted with in the sponsor hall were struggling to keep up. Even though their leadership encouraged them to embrace AI, they were only running small-scale experiments and desperately trying to keep pace with related developments.
To further illustrate this point, while the conference was running, my X/Twitter feed was alight with more developments from both companies mentioned so far. If you want to learn more about last week’s NVIDIA Blackwell chips and Microsoft’s recent acqui-hiring in the AI space, check out Ed Sim’s newsletter.
Walking away from the keynotes and three days of booth chats, I couldn’t help but think that the AI future is finally here at KubeCon—and yet, as the cliche goes, it’s just not evenly distributed; especially amongst end users. It’s clearly the time for the cloud native crowd to start building the next generation of AI-powered apps. And I, for one, welcome our new (cloud-based) AI overlords.
A call for responsible innovation i.e. “don’t forget about cost and sustainability”
Alongside the copious mentions of AI, the opening keynote called for “responsible innovation.” This phrase was used several times throughout the show. Given the context, I took this to mean “keep innovating, but do so using open source and with cost and energy savings in mind”.
As the CNCF runs the show, there are obvious calls to embrace OSS and, ideally, tech in the CNCF landscape. There was a tinge of discomfort in the audience, as Redis announced it was moving to a dual license on the same day at the opening keynote, which reminded everyone of a raft of these recent changes.
It was good to see the CNCF acknowledge some of the challenges around sustainability. The day three keynote expanded on this topic in more detail, with Gualter Barbas Baptista from Deutsche Bahn grounding the call for sustainability in a real-world (enterprise) use case, “Building IT Green: A Journey of Platforms, Data, and Developer Empowerment at Deutsche Bahn”:
This is an important topic, and it was good to see it getting plenty of keynote time.
Platform engineering takes center stage
At KubeCon EU in 2022, a few of us spoke about an emerging trend of “platform engineering”. In 2024, platform engineering has gone mainstream.
The sponsor showcase was flooded with mentions of platforms, platform engineering, and developer experience. There was also an excellent keynote session from Solomon Hykes, co-creator of Dagger and Docker, “A 10-year Detour: The Future of Application Delivery in a Containerized World” (as an aside, I chatted with Solomon on the InfoQ podcast recently and learned a lot about Dagger!)
Infrastructure layer tooling and frameworks, such as Kubernetes, service meshes, gateways, CI/CD, etc., have evolved well and are largely considered “boring tech” (at least to this audience). The big challenge now is assembling the puzzle pieces to deliver value to the internal customers: the developers. As Betty Junod astutely observed in a reply to the above Tweet, “We taken all the PaaS’es of that era apart…played with all the individual legos…got some new ones…and are now trying to reassemble them”.
I recently discussed my mental model for the layers of platform-building on the Syntasso blog. Of the three layers — application choreography, platform orchestration, and infrastructure composition — I see a potential danger of over-rotating on the application (portal) layer.
In much the same way as people are debating where the value creation and capture will occur in the AI layers, I see the same thing playing out in the platform space layers.
The infrastructure layer looks profitable but is primarily a late majority play. As Adam Jacob recently pointed out, OpenShift makes more than $1B in annual revenue. But at the same time, we are seeing market challenges for HashiCorp and Terraform (and related, OpenTOFU was very visible at the event).
I’m seeing many companies promote tooling and frameworks at the application choreography layer, typically through internal developer portals, such as Backstage, Port, Cortex, etc, and workload specification languages like Score and Radius.
My friends Abby Bangser and Whitney Lee presented an excellent talk on the topic of platform UX: “Sometimes, Lipstick Is Exactly What a Pig Needs!”
There’s clearly value to be created at the application choreography layer. However, this KubeCon appeared to be the best of times and the worst of times for Backstage. The Backstage community presented several good sessions, and there is clearly a lot of momentum (this was the project most contributed to by CNCF end users). But I was also hearing a lot of grumbles from adopters about the lack of “out of the box” functionality and the effort required to get the portal operational and maintain it.
This application layer is where the proverbial “[developer] rubber meets the [platform] road”, and arguably the developer experience is crafted here. But drawing on my software architecture experience, I’m a proponent of creating the correct abstractions, and a portal feels a necessary but not sufficient abstraction for realising the goals of a platform. I believe the platform orchestration layer is where much innovation will happen next year. I’m obviously a bit biased here, as I’m currently working with the Syntasso team (building the open source Kratix platform orchestrator), but Humanitec, Upbound (Crossplane), Massdriver, Mia-Platform, Qovery, and many others are investing in this space, too.
As a reminder of the platform-building space's relative immaturity, some of the best-attended content focused on the fundamentals of delivering cloud native apps. For example, Adrian Mouat presented “Building Container Images the Modern Way” to a packed keynote room. The Docker booth was heavily focused on Docker Build Cloud (and it was great to see the Docker folks back at the event after a year’s hiatus). And many of the 101-level sessions were full and standing.
Finally, I didn’t see AI making a big impact on platform engineering (yet!). As we’re mostly orchestrating app, platform, and infra components and managing the lifecycles, the “AI ROI” appears to be higher elsewhere.
Product thinking FTW!
In a closely related topic to the previous takeaway, I saw much more “product thinking” being applied to platforms, frameworks, and developer tooling at this KubeCon.
I thoroughly enjoyed the Platform Engineering Day colocated event and the talk, “Boosting Developer Platform Teams with Product Thinking” by Samantha Coffman at Spotify. I heard several other platform builders say this was also their favourite talk. Obviously, we can’t all be Spotify, but Samantha presented a lot of very useful mental models and made strong references to a lot of Marty Cagan’s excellent work (such as the four big risks: value, viability, feasibility, and usability):
The talk from Chris Plank at NatWest, “Unlocking Innovation: How NatWest Bank Uses Cloud Native Tools to Deliver Platform as a Product”, was also a standout presentation for me. Chris set the scene nicely for the challenges of rolling out a platform within a constrained (and well-regulated) environment and made a case for applying product thinking and cross-organisation collaboration to overcome these hurdles.
Related, Mitch Connors from Aviatrix presented “Product Market Misfit: Adventures in User Empathy”. My friend and ex-colleague, Nicki Watt from Open Credo, also presented “To K8S and Beyond — Maturing Your Platform Engineering Initiative”, which provided a useful guide to the CNCF App Delivery TAG’s Platform Maturity Model with a healthy dose of product thinking.
If the future of KubeCon is platform-shaped, I hope it also continues to be product-focused, too!
More interest in developer experience and inner and outer dev loops
I’m obviously biased here, as I presented a talk at the colocated App Developer Con, “Testing Cloud Apps: Mocks vs. Service Virtualization vs. Remocal Tools” (no video yet!). However, I saw many more mentions of the importance of focusing on developer experience and the inner and outer developer loops throughout the event.
My friends from AtomicJar (now Docker) and Diagrid, Oleg Šelajev and Alice Gibbons, presented “Simplified Inner and Outer Cloud Native Developer Loops”. I also heard a lot of chatter around the topic of cloud development environments (CDEs) from the awesome folks at GitPod. I also heard interesting things from the new kid on the CDE block, Daytona.
When speaking to senior leaders at the event, everyone appeared to know the value of developer experience, but they often had trouble “selling” this to leadership (particularly post-ZIRP). The general consensus was that focusing on DORA/SPACE metrics and cross-org collaboration can help. As I’ve shared previously on socials, you can’t go wrong by listening to a few episodes of the DX Engineering Enablement podcast with Abi Noda.
Security continues to be big business
Although the keynotes were relatively light on the topic of security, the sessions and sponsored showcase were definitely not. A quick search in the session catalogue for the term “security” showed tens of talks. Everything from secure supply chains to network intrusion and platform security was covered nicely.
Some sessions were overtly focused on security implementations and tooling, with Kyverno, Falco, and OPA stealing the show. Secure supply chains featured heavily mentioning container build tooling (alongside SBOMS and SLSA), Keptn, Harbor, etc. There was also a lot of focus on network security, with interesting mentions of Cilium (and eBPF in this context), Linkerd (with questions about Cloudflare’s Pingora), and Istio.
End-user stories are moving up the stack
Once again, it was good to see end-user keynotes and sessions on the program. I noticed that many of these sessions were “moving up the stack”. Instead of presenting how they had adopted Kubernetes (or some other CNCF tech), they framed the messaging around scaling, sustainability, cost savings, or enhanced developer productivity. The previously mentioned Deutsche Bahn keynote on sustainability was a good example of this.
I also enjoyed the Spotify talk, “How Spotify Re-Created Our Entire Backend Without Skipping a Beat” by Nick Rutigliano and Daniel de Repentigny, which applied product thinking to vending and maintaining production-ready K8s-based environments. Other good talks in this space included:
- State of Platform Maturity in the Norwegian Public Sector
- Keeping the Bricks Flowing: The LEGO Group’s Approach to Platform Engineering for Manufacturing
- Intuit’s Scaling Service Mesh: Self Service Beyond 300 Clusters
- BlackRock’s How to Save Millions Over Years Using KEDA?
- TikTok’s Edge Symphony: Scaling Beyond Boundaries with Multi-Cluster Controllers
The tooling bundling continues
As I’ve observed in previous KubeCon summaries, the bundling of tools and frameworks continued. With a few exceptions, gone are the days when vendors fought to be the “best in class point solution”.
The cloud native networking stack is a good example of this, with the booths from Isovalent (soon to be Cisco!), Solo, Tetrate, and Kong continuing to grow in size alongside the breadth of their offerings. There were several API gateway-focused vendors at the event, but even they were talking about the ease of integration with service meshes and CNI.
Building on the security bundling (by Snyk, Wiz, Aqua, etc) and observability bundling (by Datadog, Chronospere, Splunk, etc), we were seeing bundling in AI tooling from the cloud hyper scalers (see the previously mentioned KAITO) and AI-focused tooling vendors and projects like Ollama.
Wasm: A hot topic, but with some uncertainty
Wasm was once again front and center at the event. I didn’t mention this technology in my KubeCon NA 2023 summary, and several folks commented on this. This was definitely an oversight on my part. At the Paris event, there was also a lot of chatter about the AI and sustainability use cases with Wasm, and the SpinKube project received some good keynote time:
In the interest of balancing the hype, many folks I chatted to in my (platform engineering-focused) circle were still looking for the “killer use case” for Wasm. Anecdotally, the combination of container-based and serverless apps provided the required granularity and control of resource usage. They were also more interested in exploring the AI tooling when they returned to the office.
I’m generally positive about Wasm — and having worked a lot with proxies and API gateways in the past, I could always see the plugin use case with Wasm being a replacement for something like Lua — but I wonder if we’ll see less keynote time and more behind-the-scenes usage in the future.
Dapr is the cloud native ESB (in a good way)
It’s no secret that I’ve been bullish on the Dapr project for quite some time. This is a framework I wish I had access to when I was building early cloud native apps ten years ago. It was good to see many mentions of the tech throughout the event, even in sessions I didn’t expect, and the focus was often on the solid integration with other CNCF tooling.
As usual, I had great discussions with Bilgin, Mauricio, and Mark from Diagrid. We chatted about the developer patterns, the future of cloud native application middleware, and “Cloud-Computing in the Post-Serverless Era: Current Trends and beyond”.
More than one person jokingly referred to Dapr in the context of old-school enterprise service buses, framing Dapr as “an ESB for the cloud generation” or “ESB as it should have been”. Having worked with several ESBs in my past life as an enterprise Java developer, I had to both chuckle and nod my head at this! Hopefully, the new generation of developers won’t remember the jokes we used to make about ESBs.
Bonus: GTM takeaways for KubeCon EU
In my day job, I advise several developer tooling companies on go-to-market, product marketing, and developer relations, and so I wanted to share a paragraph or two on this. Looking around the sponsor showcase highlighted several trends for what’s currently hot and sellable:
- AI-based anything
- Platform engineering
- Security
- Observability
- Cost savings (and, interestingly, the “FinOps” label was not consistently applied)
This was also the first time that I noticed several KubeCon regulars not running a booth, particularly those in the VC-backed series A/B funding space. Chatting to founders and GTM folks revealed a combination of cost saving, not targeting the EU market, and believing that the lead quality was low for their ICP (i.e. booth scans from previous KubeCons had not converted to $$$/€€€).
This is anecdotal, but I also noticed many more salespeople at this event. When I started attending conferences early in my career in the Java space (in the late 2000s), it was completely normal to be accosted by overly-keen sales folks in the sponsor showcase. This gradually changed as software tooling became more product-led, buying power switched to developers, and developer relations roles emerged (see Stephen O’Grady’s “The New Kingmakers: How Developers Conquered the World”).
KubeCon booths have historically seen developer relations, technical go-to-market, and founders front and centre. These folks were present here, but I also got metaphorically (and sometimes physically) jumped by salespeople as I walked from meeting to meeting in the sponsor showcase. One salesperson was borderline hostile when I replied that I was busy and not interested in the sales pitch — so much so that I had to check my response and hold my tongue (he completely caught me off guard with several unfounded accusations, but the code of conduct goes both ways!)
I know that times are tricky and sales quotas are challenging, but you must understand your audience's culture. I don’t think the hard sell works in the KubeCon sponsor hall.
I’m personally bullish on leaning into the Technical SDR role for booth duty, which I first learned about on Shomik Ghosh’s Software Snack Bites podcast. I also believe developer relations folks will lean more into product-focused roles that will lend nicely to booth work. I’ve written more about this here:
Wrapping up: Community, community, community!
As with every KubeCon, I was impressed by the community’s efforts to make this event more inclusive. I’ll tip my hat to the Deaf & Hard of Hearing Working Group, which secured sign language interpreters for the keynotes and many other sessions.
Nikhita Raghunath and Aparna Subramanian did an excellent job co-chairing the event. I’ll also shout out my cloud native buddy Kasper Borg Nissen, who co-chaired the event for the first time! Further building on the topic of inclusion, Kasper also ran the keynote panel discussion “Unity in Diversity: A Decade of Inclusive Growth in the Cloud Native Community”.
I thoroughly enjoyed my time at KubeCon EU. It was great to catch up with so many of the community, and I had a lot of fun with a “Mastering API Architecture” book signing at the O’Reilly booth. It was also fun to spend some time with folks I’m currently working with, such as the Syntasso crew, who were co-organisers of PlatEngDay. I’m looking forward to the next evolution of this colocated event!
I’m still unpacking all of my learning, and so I’m sure I’ve missed some important topics or mentions of key people. If so, please let me know!
If we didn’t get a chance to cross paths, please do contact me via socials or the usual channels. See you in Salt Lake City in November!
