Agents in Production panel featuring @xanamini, Ron Haberman, @tibbetts, Daniel Porrras Reyes.
Glad to see infrastructure and SRE in the conversation. Valuable takeaway: Taking a stand on "what good means" as a product discipline.
Thanks @RobertBlumofe@flybridge@chazard and @Orrick@fidelityprivate and @redhat Open Accelerator for hosting. And thanks @Akamai - hope to see more events at the beautiful HQ.
The Era of Private Power ⚡️ https://t.co/fPzEokOLTA
@CyanVentures
"The best part of this moment is that we no longer need favorable policy or subsidies to drive incredible outcomes..."
I'm curious what risks/barriers could come from "unfavorable policies"
HTML is the new markdown.
I've stopped writing markdown files for almost everything and switched to using Claude Code to generate HTML for me. This is why.
We've been working with IoT systems for over a decade. One thing we've consistently seen is that these workloads generate a large amount of operational artifacts, and that data often dwarfs the system-of-record data.
For example, a device might have a few dozen configuration fields. But once it's deployed, it starts producing a constant stream of sensor telemetry, events, and other signals. Over time, that telemetry can become 100x larger than the primary dataset.
AI systems seem to be developing a similar shape.
A single interaction might generate prompts, retrieved context, embeddings, model outputs, traces, and evaluation results.
None of these are the core application data. When we talk about the "system of record," this isn’t it.
Instead, this generated data represents operational artifacts of the AI system.
Unlike traditional software logs or traces, these artifacts are often stored, analyzed, and fed back into the system.
Over time, AI systems accumulate a large, append-heavy dataset describing how the system behaves.
In that sense, AI systems start to resemble physical systems more than traditional software workloads: small amounts of primary state surrounded by a much larger stream of observations.