FIELD NOTES
Practical notes on better systems.
Token cost, AI workflows, HubSpot, and the handoffs between tools.
Token cost
How do you reduce AI token costs without lowering quality?
Start by measuring cost per successful task, then reduce repeated context, unnecessary output, and oversized model calls. Validate every change against representative evals.
AI workflows
When should you add model routing to an AI workflow?
Add model routing when your workflow contains tasks with different complexity or quality requirements. First establish an eval set so the routing decision can be tested.
HubSpot
What should you fix in HubSpot before adding AI?
Fix lifecycle definitions, required fields, record ownership, and duplicate handling before adding AI to HubSpot. Automation built on inconsistent records spreads the inconsistency faster.
Stack & integration
How do you find the owner of a broken integration?
Trace the data from source to destination, then assign an owner for the handoff rather than only the individual tools. Ownership needs to include failures, retries, and reconciliation.