AI should augment employees, not just replace them.
A good AI system expands the surface area of good judgment. If it only removes humans from the loop, it often removes the part that understood the work.
05
Principles
A manifesto for useful AI work.
The point is not to worship tools. The point is to make better workflows, products, decisions, and operating loops.
Useful systems beat impressive demos.
Demo energy is cheap. The hard part is the quiet plumbing: permissions, context, failure modes, feedback loops, and the boring day-three workflow.
Context engineering matters more than prompt tricks.
The best prompt cannot compensate for missing knowledge, unclear boundaries, broken data, or a process nobody has bothered to describe.
Governance without adoption is theatre.
Policies are useful only when they shape how work happens. Otherwise they become a folder people cite after the risky thing already happened.
Agents are workflows with judgment, not magic pixie dust.
An agent is a loop with tools, memory, constraints, and escalation paths. The magic is deciding what should be automated and what should stay visible.
Ship small, learn fast, compound.
A small working system tells the truth faster than a 40-slide fantasy deck. Build the smallest useful thing, then let reality improve the plan.