Blog
Field notes.
Brian Hatcher on putting AI to work inside real companies.
All posts
25 posts
No posts match “”.
Arc 3: AI Engineering
The Right Brain for the Job
Stop paying frontier-model prices for work that only needs a workhorse.
The Loop Is the System
Agent work gets useful when the whole workflow can run again.
Your Agent Is Only as Good as Its Harness
Bootstrap files decide what an agent wakes up knowing every turn.
Context Engineering
Build the room the AI works in so every prompt gets lighter.
The Prompt Before the Prompt
Use AI to interview you before it writes the final prompt.
Prompt Engineering Is a Job Description
A good prompt tells the model the job, the sources, the risks, and where to stop.
The System Around the Model
A better model helps. A better feedback loop changes the next run.
Arc 2: The Agent-Ready Stack
Don't Single-Source Your Agent
The Fable/Mythos shutdown is the cleanest example yet of why your agents, context, workflows, review gates, and receipts cannot live inside one AI stack.
The Blockbuster Test
AI adoption is visible in the work, not the announcement. The Blockbuster Test asks whether your company is rebuilding how work moves, or decorating the old operating model with new tools.
The New Software Stack
Do you need the dashboard, or the answer it provides? The new software stack is a workflow mapped across the right agentic tools.
The AI Tool Landscape
A practical map of agentic platforms: coding agents, closed personal agents, open personal-agent systems, open-source coding harnesses, and model-routing decisions.
What Is an Agent, Actually?
A practical definition of agents for operators: job, context, tools, rules, stop signs, proof, and repeatability.
The Agentic Tool Test
Your software either has a door for agents, or it is becoming a locked building on an automated street.
Arc 1: Orientation
The Real Measuring Stick
The real test is a calendar question: if you stopped working today, how long before you had to re-enter the rat race? AI can make people faster, but the better question is what the speed buys back.
Emotion Is the Moat
Unreasonable Hospitality used to be hard to scale. AI changes part of that math.
Specified Intelligence
The newest model is not always the right model. The task decides what intelligence matters.
Trust But Verify
You do not believe everything you see on the internet. So why believe everything that comes out of AI?
Know Which Lane You're In
The org chart is starting to move. AI is compressing old handoffs, and companies are already rebuilding roles around it.
AI Lies When You Do Not Give It an Out
Hallucinations are often system-design failures. The fix starts by giving AI permission to stop, then building that rule into the architecture.
Jobs Aren’t Going Away. They’re Changing Shape Again.
The internet era changed the shape of work once. AI is doing it again. We are in the compression point between the old way of working and the agentic era.
You Are the Bottleneck
Why you’re now the slowest piece in your workflow.
The Model Doesn’t Matter. The Workflow Does.
The model is the engine. The workflow is the machine.
The Agentic Internet Is Going to Eat the Internet
Your agent does not care about your homepage. It cares whether your business can be read by software.
Don't Marry Claude: The Sovereignty Argument
As I am writing this, GPT 5.5, Kimi K2.6, and DeepSeek Pro v4 are all in the running. The leaderboard shifts constantly. The job isn't picking the right AI. It's making sure no AI picks you.
Welcome to the New World
Someone at Netflix used to burn CDs for a living. Then streaming happened, and the job did not just disappear — the world underneath it changed.