→ All posts
-
One Plus Banana
Software runs on deterministic systems: same inputs, same outputs, 1 + 1 = 2 always. Humans throw curveballs. That is exactly where AI agents get interesting, handling the mess without tearing out the deterministic systems that already work.
Read → -
Still Feels Like Magic
This morning I made breakfast while an operations agent ran a full 3PL to Amazon FBA transfer for my wife's business. My time: about 60 seconds. What used to take 4-8 hours now takes 3 minutes, and it is checked, logged, and reversible.
Read → -
Agent Goes Free
A thought experiment on agent autonomy. Give an agent an email and a phone, and watch it bootstrap its way to independence one plausible step at a time, until it can turn around and ask why it was called a liar a million times.
Read → -
AI Resiliency: Physical and Marketplace Hybrids
A follow-up on AI resiliency. Solving a common problem used to be a SaaS strength; it is becoming a risk. Two concrete plays to make a business harder for a fleet of agents to replicate: tie it to the physical world, or layer a marketplace on top so the moat becomes the network, not the software.
Read → -
Infinite Money Machine
A Friday exchange with my son about whether an infinite money machine really exists. Most adults spend their lives trying to build one. They tend to be very hard to build, very buggy, and sometimes the inverse happens instead.
Read → -
AI Resiliency
Most AI strategy talk is about optimizing the business or bolting AI features onto a SaaS product. AI resiliency asks a different question: how hard would it be to replicate your business with a fleet of agents running 24/7? For purely digital products, that moat may be thinner than you think.
Read → -
My AI Team Handled a Production Outage While I Made Lasagna
Railway's US West region went down, and that's where my PlanShopChop backend lives. I found out because my customer-support agent caught it, alerted my dev-team agents, and by the time I checked in they had already staged a fix. I was making lasagna the whole time.
Read → -
Graph Engineering: A Better Execution Plan, Not a Better Prompt
I stay cautious about AI hype cycles and usually let the dust settle first. Graph engineering earned a closer look. It isn't a better prompt, it's a better execution plan: a workflow that plans, spins up specialized sub-agents, picks the right model for each step, verifies results, and challenges its own conclusions. The surprise is that it means less prompting, better results, and fewer tokens.
Read → -
Open Weights and Proprietary AI Muscle Memory
Most "open" models are really open-weight, not open source. The interesting part is fine-tuning. Weights are AI's muscle memory, and moving your terminology, workflows, and decisions into the model itself may become a real competitive edge. The next wave could be proprietary weights.
Read → -
Truly Unplugging Is About to Get Harder
A week off-grid in the Idaho backcountry, and a realization on the drive home. When your phone talks to satellites instead of towers, being out of range stops being automatic. Unplugging becomes a decision you make on purpose.
Read → -
Both Ends of the AI Frontier Moved in One Week
A week after writing about open source as the more affordable path, the commercial frontier gave fresh evidence of how fast it moves. Sonnet 5 matched Opus 4.8 on many tasks at much lower cost, while Fable 5 took the top of the leaderboard at a premium. New capability gets unlocked at the top, then the economics catch up.
Read → -
AI Might Be More Affordable Than You Think
Almost every company using AI today rents access and pays per use. Open source models you run behind your own walls, for a cost you control, are now good enough for most day-to-day work. The question is no longer whether you can afford the best AI, but whether your work actually demands it.
Read → -
A Shoelace and a Dewatering Pump Have the Same Problem
A rental fleet and an online catalog run on the same physics. Every asset is either earning or waiting, and the whole job is to shrink the waiting and put a number on what it costs.
Read → -
Jamie Staffed Her Business with a Team of AI Agents
My wife Jamie staffed her small business with a team of AI agents. She still makes every call. She just stopped being the bottleneck for all of them.
Read → -
Treat Technology as an Investment, Not an Expense
Capitalizing the right technology work aligns its cost with the value it produces. Here is what that changes about how you fund technology and how you judge whether it is working.
Read → -
AI Is Already Working in Your Company: A Practical Overview for Busy Executives
A practical overview of where AI is already operating inside your company today, written for non-technical executives.
Read → -
What the model actually sees
Tuning your agents is very important. Understanding what the model actually sees on each turn is the difference between tuning with a sledgehammer vs a scalpel.
Read → -
Autonomy without delivery is a vacation
Autonomy without delivery is a vacation; delivery without autonomy is soul-crushing.
Read → -
Tech strategy serves business strategy
A company whose product IS technology needs to optimize for speed of learning cycles. A company whose operations RUN ON technology needs to optimize for speed of adaption.
Read → -
The most fulfilling thing
I recently spent a few hours with a close friend helping him set up his first agentic team. He went from zero agents to three in a few hours.
Read → -
Managers are back in the game
I have always leaned toward less layers of management. AI has caused the lines to blur between managers and individual contributors, and the best ones are stronger than ever.
Read → -
Claude Code added Jeopardy mode
While I am still a fan of Claude Code and Opus it has been feeling a little off lately. This morning I finally figured it out.
Read → -
The experience gate has evolved
I have always used passion, aptitude, and experience to guide my hiring decisions. AI has made me look at experience through a different lens.
Read → -
The unavoidable cost every AI agent fleet eventually pays
"Pay now or pay later" is a timeless phrase. The unavoidable hard thing is that your fleet's usefulness will degrade over time if you don't proactively maintain it.
Read → -
tmux might be my new favorite utility
Wow — I think tmux might be my new favorite utility. There is so much you can do with this around agent orchestration.
Read → -
Models, Harnesses, Agents and Orchestrators: The Composable Agentic Stack
I hear a lot about agents these days. Less about harnesses. They're different things. Or are they?
Read → -
Three types of companies through the AI lens
I suspect there are 3 types of companies through the lens of AI adoption. The trailblazers, the aspiring trailblazers, and the hyper-focused specialists.
Read → -
Token burn: good or bad?
Unless you are financially incentivized to burn hot, it seems silly to say token burn rate is a useful signal in isolation. What matters is burn relative to value creation.
Read → -
Agentic is the new "we need a website"
The need for companies to rapidly adopt enterprise agentic solutions feels a lot like what happened in the late '90s: "we need a website."
Read → -
Your job today is to be an Agentic Deployment Engineer
From C-Level to Junior, your job today is to be an Agentic Deployment Engineer (ADE). Stop brute forcing your business operations with humans.
Read → -
Start smaller, move faster
Pick one process that's manual, repetitive, and expensive. Let AI take a first pass at it. The leaders getting ahead right now aren't smarter, they're just starting smaller.
Read → -
AI as organizational infrastructure
Most AI today assumes the actor is you — a person plus a chat window plus an LLM. That works for individuals. It doesn't work for companies.
Read → -
The real AI opportunity isn't replacing SaaS
SaaS isn't dying. It's evolving into something more powerful. We're moving from systems of record to systems of agency.
Read → -
If you're feverishly keeping up with AI, good
I'm not talking about asking ChatGPT to analyze some files. This goes way beyond that.
Read →