Technical
My 2026 Thesis: Agents Become Infrastructure
Every founder I know is trying to figure out where AI fits in 2026. I spent the holidays rewriting my own playbook. The thesis I am betting the next twelve months on is simple: agents stop being novelty and become infrastructure. Here is what that means for a solo consulting practice like mine and what I am shipping because of it.
From Demo to Dependency
In 2025 I used agents because they were fun. In 2026 I use them because my workflows break without them. That is the shift from experiment to infrastructure. When a tool moves into infrastructure, three things change:
- You pay for uptime, not novelty
- You version and monitor it
- You refuse to ship without it
The same pattern played out with databases, then cloud, then CI. Every layer that crossed from novelty into infrastructure eventually rewrote the expectations of the layer above it. Agents are crossing that threshold right now.
What I Am Building Differently
Every new client project now assumes an agent is part of the stack. Not as a chatbot bolted on, but as a pipeline component. Batch classification, content generation, internal search, triage: these used to be hand-written rules. They are now small agent calls with strong tests around them.
# 2025 pattern: rules-first, agent-optional
# 2026 pattern: agent-first, rules for guardrailsThat ordering matters. Rules become the safety net, not the main logic. The agent handles the gray zone and the rules catch the edges. It produces less code overall and the code that exists is easier to reason about.
The Bets I Am Making
- Claude Code stays my primary builder
- Model routing becomes table stakes (haiku for volume, opus for judgment)
- Every client deliverable ships with an internal agent for the team
- Eval-driven development replaces test-driven development at the agent boundary
- Prompt files become versioned artifacts, tracked like code
I wrote more about this direction in the Anthropic API overview which is the foundation I build on.
The Risk
Infrastructure bets mean switching costs. If the bet is wrong, the rewrite is painful. I am comfortable with that because the alternative, treating agents as permanent experiments, is slower and more expensive. Every month I delayed this commitment last year cost me speed I cannot recover.
What You Should Decide
Pick your 2026 thesis by January 15. Write it down. Re-read it every Friday. That single habit separates founders who ship from founders who drift. A thesis is not a prediction. It is a constraint that tells you which opportunities to say yes to and which to let pass. Without that constraint, every new tool looks interesting and you end up with a workshop full of half-finished experiments.
RELATED READING
The Consulting Shift I Am Making In Year Two
After a year of writing and building, my consulting practice is changing shape. Shorter engagements. Sharper outcomes.
ReadThe Frontend Shift: Shipping Less JavaScript In Year Two
A year ago I reached for Next.js for everything. This year I often reach for nothing.
ReadThe Serverless Lesson I Would Write On A Sticky Note
After a year of shipping serverless projects, one rule explains most of the wins and all of the losses.
Read