Joshua 1:6 Be strong and courageous, because you will lead these people to inherit the land I swore to their ancestors to give them.

It’s been a while since I’ve sent a leg room selfie. Scored Southwest First Class!

What do you want to know?.

Where am I headed?

Just moved my RAG pipeline from a custom fine-tuned model + vector search to a foundation model grounded in Vertex AI Search. Yes, I lose some of the finely tuned grounding….But I also lose $1,700/month in fixed cost. #VirtualCTOAdvisor.

I've learned more about building RAG pipelines than I've ever wanted to know. I have to re-engineer my complete pipeline based on my learnings.

Found the runaway charge on @googlecloud for #VirtualCTOAdvisor 💸. Turns out:.👉 Fine-tuned models require an always-on endpoint.👉 That means a VM that doesn’t scale to zero.👉 Just sitting there, quietly burning cash 🔥. The fix?.➡️ Switch to a public Gemini model = drops.

🚨 Mid-cycle update on the @GoogleCloud bill for #VirtualCTOAdvisor:. 💰 Vertex AI: $273.22.💡 Everything else: $2.23 combined.📅 7 days still left in the billing cycle.Even with GCP credits, this is a Buy vs. Build moment. Vertex AI, as I have it configured, is too expensive for

No @Veeam or @Cohesity sponsorship of VMware Explore this year. Interesting as a huge part of their target customers are VMware shops.

VMware Explore has about 47 sponsors. Not too bad given the disruption in the partner echo system. AWS, GCP, and Microsoft are all Diamond sponsors. IBM is platinum and no Oracle.

The scream you heard in the data center was me shelling out $700 to @Broadcom for #VMwareExplore.

AI didn’t replace the junior dev. It became one. Example: Cursor pulling data from logs instead of querying the actual DB. The data's right there—but AI doesn’t yet grasp context, just patterns. It’s useful. But far from the “autonomous engineer” narrative we keep hearing.

Gemini called me an analyst.

I’ve been creating more content than my audiences can consume—and that’s a byproduct of my intentions. It’s part of preparing for #VirtualCTOAdvisor. A proper digital twin needs data. The more data, the more accurate the digital twin. Ironically, the future of my content.

Bad habit I picked up from @danielnewmanUV

I'm really liking Cursor AI. My complaint? It's on my MacBook. It's essentially my SRE Agent. What I want is Cursor AI as a service. So that I can run it from any machine because it's essentially my SRE. @nathaniel_avery, do any of the products you manage give me this capability?.

The bots on here have been using AI and have gotten really good. This one is in response to my last post complaining about public AI models. If not for the pitch for some Crypto account, I might have engaged. The user handle isn't even bad.

Chat GPT-5 shows the risks of using a public model for production work flows. The change in quality has changed my writing workflow to the point I’m testing Gemini. I’ll probably move this process to a local model over the longer term.

I wonder how many “I love Broadcom” stickers I’m going to see at VMware Explore next week.

What’s the end game of the AI training race? How do you protect your business from the winner?. These Super Intelligence and AGI models will be trained and fine-tuned by companies with resources far greater than the people consuming the models. Yes, we’ll be able to locally run.