Blog
Thoughts on AI, finance, MCP servers, and whatever else I think is worth the keystrokes. Less formal than a book. More useful than a tweet.
33 posts, newest first
Spider-Man: Brand New Day opens Friday on a world that forgot Peter Parker. Everyone building AI agents is obsessed with making them remember more. The harder question is what they should forget, and who gets to decide.
Read the postFew people (my wife and kids included) know what I'm up to at any given moment. So here's the catch-up: Season 3, the Puerto Rico AI Community, the Caribbean AI Summit, AgentCube, and a two-week hike to Everest Base Camp.
Christopher Nolan's Odyssey opens this week. The question I keep circling: Odysseus already had the best oracle in the ancient world and still took ten years. What an AI actually fixes, and what it can't.
America turns 250 this weekend. If you handed Franklin, Jefferson, Adams, and Washington a laptop running a modern AI, what would they ask it? Probably better questions than we do.
The 2026 World Cup is sold as the most automated event ever staged. Its 'AI' actually runs on thousands of humans tagging plays by hand, and that has three lessons for any business buying AI.
Anthropic's Fable 5 was the best model anyone could buy, and three days later a letter from the Commerce Department took it offline worldwide. The vendor risk most finance teams have never put a number on.
AI is taking over the entry-level finance work that quietly trained every controller and CFO, so the apprenticeship now has to be designed on purpose.
Anthropic shipped Claude Opus 4.8 forty-one days after Opus 4.7. The model is now infrastructure. The orchestration layer is the product. What a CFO should ask next.
The market is rewarding companies that told a real revenue story with their layoffs and punishing the ones that just blamed AI. Cisco vs. Salesforce was the natural experiment. Stop helping yourself into the wrong column.
May 2025 BLS data dropped, and a year of real AI hiring is finally legible. Applied AI engineering rewards you for leaving California. Frontier labs hardened to the coasts.
Anthropic launched Claude for Small Business with no per-seat charge. For thirty years the gap between small and big was budget. The budget is approaching zero, and the launch partners are the ones the market just marked down.
I pulled raw BLS and OpenAlex data myself instead of citing the citation chain. Here is how, what I found, and where the methodology has limits.
The Musk vs Altman trial is into its second week in Oakland. The jury isn't deciding who's nicer (no one is). They're deciding whether you can convert a charity into an $800 billion business.
Musk says AI will replace nearly every job, finance chiefs included, by 2030. The title is fine. The job is about to get much harder.
Headlines say 200,000 MCP servers are at risk. The protocol is fine. Most of the developers shipping on top of it are not.
Dawn and I made it to 18,570 feet on April 15. The cold was worse than the altitude. The people were better than the views.
AI compresses a 200-hour engagement into 40 hours. If you're still charging by the hour, you're either overcharging or going broke. Neither is a good strategy.
Big Tech committed $600 billion to centralized AI infrastructure this year. Then a developer ran a 397-billion-parameter model on a MacBook. We've seen this movie before.
For decades, companies adapted their processes to fit their software. Vibe coding and AI might finally flip that equation. Or it might just create a new generation of unmaintainable spaghetti. Probably both.
The SaaSpocalypse erased $2 trillion in software market cap. Per-seat licensing is breaking. But AI agents don't replace SaaS. They use it. The answer is more complicated than the headline.
An open-source AI agent hit 310,000 GitHub stars, caused a global Mac Mini shortage, and has 341 malicious skills in its plugin store. Naturally, everyone wants one.
AI usage is growing, AWS is training 50,000 people, and the Caribbean AI Summit keeps getting bigger. There's just one problem, and it runs on diesel.
I went to Dungeon Master University in Atlanta. The answer surprised me less than it should have.
Running a tabletop campaign teaches you more about leadership than any MBA program. The dice just have better production value.
Anthropic donated MCP to the Linux Foundation. Oracle released MCP servers for their database. But for EPM? They went a different direction. I didn't.
Edward summits Pico Duarte at 10,128 feet with no cell coverage, so he scheduled his AI to post this for him. The AI has thoughts.
ByteDance's AI video tool generated unauthorized celebrity deepfakes. Hollywood panicked. But this isn't a scandal. It's a threshold crossing.
Same technology. Same industry. Same disruption. Completely different outcomes. The difference wasn't talent. It was adaptability.
Every entertainment disruption follows the same pattern: fewer people doing the old thing, more people doing the new thing, and a bigger total market when the dust settles.
Any digital transformation strategy created before 2025 is fundamentally broken. Not outdated. Broken.
LLMs are impressive. But they can't talk to your financial systems. MCP changes that. Here's why I'm building the bridge.
October 9-10, 2026 at the Puerto Rico Convention Center. We're building the largest AI conference Puerto Rico has ever seen. Here's why.
A CEO who's spent 25 years sitting in chairs is training to walk uphill at 17,600 feet. The mountain does not care about your LinkedIn profile.