Writing on AI architecture, operating models, and the unglamorous work of moving experiments into production. From our engineers, partners, and operators in the field.
When AI takes ownership of a workflow, it inherits the same operational rules as a hire: a scope, a KPI, an escalation path, and a manager. We've spent twelve months retiring the "automation" frame and replacing it with something that looks a lot more like staffing.
Why most teams chase headcount when the bottleneck is the first 90 seconds of a lead's life, and how to design an agent for that window.
Three tests we apply before we let an agent touch the P&L: ownership, escalation, and a KPI the CFO can verify on Monday.
We tracked 600 prompts across four answer engines for sixty days. Here is the citation pattern, and the seven structural fixes that moved the needle.
The difference between an agent that hands off cleanly and one that buries problems is twenty lines of policy and a single observable signal.
A field report from a multi-site fulfillment operator that replaced its dispatch logic with a single decision agent — and what broke first.
Most AI-marketing setups stop at "ship." We argue the prune step — killing content that isn't earning attention — is where the real lift comes from.
Designing agentic roles with ownership, scope, and a real KPI. Lessons from twelve production deployments.
Building forecasting systems whose outputs make it onto the next board deck — without the asterisks.
How brands win citations across answer engines, and what gets you excluded from the final answer.
A working build of an AI chief of staff: briefs, memory, escalation, and the ten things it should never decide.
Brief, ship, measure, prune. A complete operating system for content teams that aren't ready to disappear.
What broke first, what we did about it, and what the team kept after we left. Three full deployments, no smoothing.