Lab · 1
Can AI replace a CMS?
Dr. Paul tests what happens when AI takes over much of the traditional website production workflow.
Hypothesis
AI can eliminate much of the production layer when given the right constraints.
Result
Execution wasn’t the constraint. Judgment was — this page is itself running evidence.
Single-subject test, still running — not a controlled study. Read this as one practitioner’s evidence, not a general claim.
Method
Brief
What should exist, and why.
AI
Builds it — code, copy, and design.
Human judgment
Reviews, edits, decides what ships.
Live
Deployed automatically, in about a minute.
Tech
The experiment
The artifact isn’t a writeup — it’s this site, live right now. Every page was produced through the pipeline above.
AI handled
- Code
- Iteration
- Troubleshooting
- Deployment
Dr. Paul handled
- Direction
- Judgment
- Taste
- Final decisions
Worked: AI handled the implementation surprisingly well.
Constraint: Context matters. Without enough context, AI can drift.
Evidence
The real numbers behind this page
Pulled straight from this repository’s own commit history — nothing estimated, nothing invented.
Input
162
commits
9,892
lines added (net +6,483)
8 days
elapsed since the first commit
Output
8
live pages shipped
~1 min
commit-to-live deploy time
4
old URLs preserved as permanent redirects
As of this writing — these numbers grow as the experiment continues.
Finding
Execution wasn’t the constraint.
Judgment was.
AI could quickly build what was requested. The harder question was whether it was worth building.
Conclusion
Execution became more agile and cheaper.
Human judgment became more valuable.
AI didn’t replace the marketer. It changed where the marketer adds value.
Executive implication
Don’t just add AI to the workflow. Redesign the workflow around AI.
The opportunity isn’t simply replacing marketing technology with AI. It’s redesigning the marketing workflow around what humans and AI each do best.
Next Question · 2
Should Dr. Paul become an avatar?
Talking avatar, text chatbot, or shelved entirely — that’s the question this finding raised.
Back to the Lab