Field Notes
Working notes on
the bleed.
What we keep running into in the field: why legibility comes before AI, what an AI-native company actually looks like, and the small structural ideas that decide whether agents help or flail.
Why AI transformation is mostly not about AI
Before a company can become AI-native, it has to become legible. A field note from Ryan Van Sickle.
The legible organization
Documentation describes the company you meant to build. Legibility describes the one you actually run.
The AI-native company won't look like a SaaS company with copilots
Most "AI-native" claims are a SaaS company with an LLM bolted on. The real thing is structurally different. The bet is structural, not tool-stack.
Your company needs a world model, not another app
The missing layer in most AI rollouts isn't a workflow tool. It's a representation of what the organization is, knows, owns, permits, and intends.
You don't automate roles. You transfer mandates.
You don't hand an agent a job title. You hand it a specific mandate, with the authority and the limits attached.
A content agent doesn't need a better prompt. It needs a canon.
A knowledge base stores information. A canon establishes authority. Most companies have the first and assume it does the work of the second.
Agent memory is not one thing. It's five.
Treating agent memory as a single thing is why agents forget what matters and cling to what doesn't. It's five distinct systems.
Your company has four different things and calls them all "goals"
KPIs, objectives, OKRs, milestones: four structurally different commitments collapsed into one word. The collapse is why goal-setting feels like theater.
AI-native people aren't power users. They're different thinkers.
The skill-badge framing misses it. AI-native people develop different instincts about what to externalize and what not to delegate.