5 lessons from Nvidia, Kubecon & RSA

Hey there! Welcome to Platform Weekly. Your weekly print of the platform engineering boarding pass. Every week, we dive into best practices, news, lessons, and updates from the platform engineering community.

P.S. Check out the new open source AI agent harness, AURA. It’s a production-grade runtime + orchestration layer that lets AI agents run as real systems (with workflows, tools, observability, etc.).

5 lessons from Nvidia, Kubecon & RSAC

We’ve had one of the most hectic few weeks in Platform Engineering Community history - 6 in-person events, 5 conferences, and 2 talks across 3 countries.

In Amsterdam, I got the chance to take the full temperature of EMEA at our roundtable, workshops, and Exec dinner in Amsterdam. While at the same time, we scouted out the world at Nvidia GTC in San Jose, Qcon in London, RSAC in San Francisco, and SREcon in Seattle. Here are my 5 crucial takeaways from this platform engineering world tour:

Identity management is the new perimeter: It couldn’t be better timed to share our latest course on Infrastructure Identity for Platform Engineering. It was clear, diving into the pandemonium of RSAC, the world’s largest cybersecurity conference, that with the explosion of machine-to-machine interactions (API keys, service accounts, AI agents, etc.), NHI management is on the path to top priority

Cognitive load is going up, not down: New survey day (plus anecdotally from dozens of convos over the last 2 weeks), 2/3 of engineers say they are more stressed than 5 years ago with the onset of AI. Overwhelmingly, as AI speeds up throughput through code gen, research, and prototyping, most of those I talked to are just ending up with more parallel work, more unfinished threads, and more decisions per hour - and nowhere near less work. 

Throughput is massively up, but nothing else is: New data confirms that throughput is up 59%+ on average, but only 5% are actually turning that into real delivery improvement. For others, despite higher throughput, they’re seeing more changes failing on main, worse MTTR, and, for the most part, are running faster and faster but staying in the same place.

The AI bubble isn’t going anywhere: Each of these 5 conferences focuses on different domains, but to no one's surprise, 95% of what every booth is talking about, every talk is exploring, every Meetup, meeting, and coffee chat is zoomed in on is, of course… AI. But not the way we used to. Even just 6 months ago, we still dealt overwhelmingly in abstracts and hypotheticals. Now I’ve seen dozens of separate demos of coding agents, watched AI SREs diagnose and remediate issues that would’ve taken hours to otherwise solve, and watched one of the more advanced examples, an agent control plane managing half a dozen agents rebuilding a legacy monolith while the rest of the team focuses on new apps. What made these systems actually work wasn’t just the models; it was the platform scaffolding underneath them - identity, boundaries, and validation. We are moving from the messy middle to build real, long-lasting, and high-powered agentic development. Stay tuned for a lot more from me on this topic.

Every layer of AI needs platform engineering: Whether you’re talking about AI-native infrastructure and GPU management with Nvidia, AI SRE best practices in SREcon, or managing a fleet of parallel coding agents at KubeCon… It’s the teams with platform engineering who are both pushing ahead into new frontiers and actually delivering on the promise of these core AI trends.

What were your takeaways? Reply and let me know.

Quick bites

Highlight of the week:

• AURA!! Go check it out.  If you're thinking about how AI actually runs in real environments, not just demos, this is a practical look at where the space is heading (and for some of us, we are already there).

From the community:

• Want to watch the recordings of PlatformCon workshops and talks you weren’t able to attend? We’re uploading more every day on the community Youtube channel!