Keith Townsend

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.

@engineering_bae We tease him about it all the time.

@CTOAdvisor 🤣 I'm glad you're saying "was"

Dude came into the gym doused in Axe cologne. And while that in itself isn’t a crime, I kind of think it should be 😭 Immediate headache.

We're two months post-Something Big is Happening... Are we on pace of to replace all knowledge work with AI? x.com/CTOAdvisor/sta…

@MichaelDell If you hadn't seen my earlier post. I don't think the shape of AI compute is in giant compute clusters. It's in the AI PC which the market was too early on. More memory, more integrated accelerators, and more connectivity.

PC & smartphone demand was always bounded by one thing: people. But what if billions (or more) of agentic AIs aren’t bounded by people at all? What happens when they never sleep… …when they work, talk, design, buy, sell, and decide 24/7? What then happens to compute demand?

There are Three Types of CTOs 1 - The Ineffective One 2 - The Risk Taker 3 - The One that Gets S--t Done Each one follows my content. Only one gets invited to a Buyer Room. Over the years, I’ve recognized the pattern. Each one is a Gartner customer. They just use the research and categories differently. Type 1 - The Job Hopper You’ve seen the resume. Every few years, this persona recycles. In 2012, they “built a private cloud.” It was a VMware cluster. They are the type to buy an AI Factory because the Magic Quadrant said that was the thing to do. They use it as cover for the fact that the private cloud was just a VMware cluster. In 2012, they cloud washed. In 2026, they AI wash. Eventually they get found out. Then they find a new job. Type 2 - Read an Article on a Flight This one is looking for anything in the upper right of the Magic Quadrant. Damn the business case. They have a resume to build, and the latest tech stack helps build it. This is the one that moved your mainframe applications to microservices in 2018. Nothing works the way it should, but the company did get a great write-up in CIO Magazine. They are now building a 10K node training cluster to solve a business problem that could be solved with a bunch of classical ML on that same mainframe. Type 3 - The Buyer Room Invite This is the hybrid. They understand that, to protect their people, they have to give plausible tribute to the Gartner selection. If the project fails, they have to be able to point to traditional market insight. But they also know that if they have to buy an AI Factory, they are going to build a Layer 2C reasoning layer on top of that infrastructure stack. This is the persona engineers will run through a brick wall for. They are politically protected and get to do cool stuff that has real business value.

I wrote something… well, actually, I co-wrote something. I applied my own Decision Authority Placement Model (DAPM) to how I use AI for writing. AI is a co-thinking partner. It helps me pressure test ideas. Where are the holes in my logic? It helps me get to a decision faster. I then take that and run with it, spitting out a stream of thought. That’s my normal style—draft → publish with very little refinement. That’s the voice I’ve built. Over the past year, this drifted. When I use AI to co-author, it breaks that voice. It goes from draft → publish to draft → polish → publish. And my voice gets lost in the process. The realization is that both systems are valid. It’s medium and intent-driven. Whitepapers, use cases, frameworks → draft → polish → publish Blog posts, Substack, LinkedIn, X → draft → publish Same tool. Different placement of authority. That’s DAPM in practice. Full case study: thectoadvisor.com/blog/2026/04/1…

I'm going to create a OpenClaw employee on a MacOS instance in AWS. I'll give it a corporate card and a prompt. My monetization plan? Adwords revenue from the resulting YouTube video about how I built a business with someone else's money and no employees using AI.

The branded AI PC was too early. It will be a thing. Look at what OpenClaw has done for Mac Mini sales.

@QuinnyPig Dude!!! You need to get better at your Route 53 query optimization. I've never had a Route 53 DB bill more than $25K!

Not true for when you accidentally get a $50k Route 53 database bill surprise, I tell you hwat. They very much want to talk to you then.

Speaking of AI... This is what ChatGPT wanted me to post. More polished but not me: I’m not well-informed enough to challenge Michael Dell on the scale of the memory explosion over the next few years. But I do challenge the shape. This assumes memory demand concentrates in large systems. I don’t think that’s what happens. If the data stays put—and it usually does—you don’t keep pulling everything back to a centralized system to make decisions. You push compute closer to where the data lives. That means more systems holding working state. More RAM, spread across more places. So yes, demand probably goes up. I just don’t think it shows up as bigger clusters.

I'm not well-informed enough to challenge @MichaelDell on the scale of the memory explosion over the next few years. But I do challenge the shape. I believe Agentic AI will drive more distributed consumption of RAM. I see this as a data gravity problem. We are going to move AI to the data, which means faster and more RAM for the client and other edge devices. wccftech.com/dell-ceo-says-…

@vmiss33 It's literally based on price. Same size models, and you don't care about CUDA, price becomes the factor.

Why a RTX 3090 over a Radeon RX 7900 RTX?

@PatrickMoorhead It's critically more important to my business than X. I don't go to LinkedIn to consume content. I go there to be the voice that's followed. It's easily the source of at least 50% of my inbound revenue.

I will admit, I do a bit of a side eye when tech companies ask me about LinkedIn posts or lack thereof. LI seems sterile and slow and old. The LinkedIn editors determine what goes viral. In-context videos are suppressed so what’s the point. The entire platform for me is low ROI with the constant out of context ads, 100 suggestions to play PUZZLE GAMES, job suggestions to be the manager of a McDonalds (see image), and constant connection requests from lead gen company founders and overseas dev body shops. My most read content is health related LoL. I do appreciate my 39K followers and will continue to post slop there.

I'm a good writer. Not a polished writer, but a good writer. Over the past two years, I let AI drift me into a polished writer. The problem is that it lost some of my voice in the process. This post is a return to my more natural style. It's evidence of me tearing down and rebuilding my swing. thectoadvisor.com/blog/2026/04/0…

A good friend: My son wants to get a degree from a top CS school in Cybersecurity. I'm trying to understand placement rates for graduates with cybersecurity degrees from that school. Me: AI is moving too fast to make any decision based on current placement rates. He is going to a top school that's constantly refreshing its curriculum. The CS degree will serve him well. Friend: But he wants to do cybersecurity. Me: He is 18. Of course, he thinks he knows what he wants to do. The point of the program is that he will narrow his focus even if it's within cybersecurity. He is getting a CS degree from one of the best schools for CS. He'll be fine.

@iamKierraD As I was reading... Me Me You lost me... Let's replace this with "Infrastructure" 😂

“Oh you do thought leadership and content creation! What’s your niche??” Me, with the most serious of faces: “Data, AI, and being hot”

I’ve been spending the morning trying to fix my AI-powered workflow. Last year, I mastered a workflow where AI was a powerful co-thinker around pressure testing ideas and drafting blog posts. Later versions of the model broke this workflow. AI became an even better co-thinker but edited my voice too heavily. However, I didn’t want to give up the throughput. So, I kicked the can down the road. This mimics how enterprise AI projects fail. I call it Decision Authority Placement Model (DAPM). In my workflow, I explicitly gave AI the authority to co-think. Where my AI-powered process broke is the implicit authority it inherited to change my writing style. That wasn’t the original system design or intention. Originally, the output was my style. Then the model changed and that misplaced authority surfaced. Now I have to go back and either fix the system with prompt engineering or put the human back into the writing portion. This post is a great example. AI helped pressure to test the idea but this post is entirely my unpolished voice.

@vNathanBennett The leader was a bit younger, and I could tell this was his first cloud migration. It speaks to something much deeper that I'm exploring this morning as we tackle AI Infrastructure.

This lives in my head every time someone says “We gotta get away from VMware”. A cloud migration is a cloud migration. Even private cloud.