For most of my career the organizing question was speed. How fast can we ship. Building was the slow, expensive part, so reducing time-to-ship was the whole game. Building got fast and cheap, so speed stopped being the thing in short supply. A different number moved to the center: the cost of being wrong.
What speed used to hide
When shipping took two quarters, the slowness was doing quiet work. A bad idea had to survive planning, scoping, building, and review before it reached a customer, and that gauntlet killed a lot of mistakes by attrition. The cost of being wrong was high in calendar terms but capped in blast radius, because wrongness traveled slowly. Compressing the build collapsed that buffer. A decision can now go from idea to in front of every customer in days. That is a gift when you are right and a liability when you are not.
Triage by reversibility, not size
The old triage was by size: big decisions got a meeting, small ones got waved through. That heuristic is now misleading, because a small decision like a default setting, a prompt, or a threshold an agent uses can be catastrophically expensive to reverse if customers build on it or an agent runs it at scale.
Use two buckets.
Cheap to reverse. You can undo it on Tuesday with no lasting damage. Most UI choices, most copy, most experiments behind a flag. Ship now and learn. The deliberation costs more than the mistake would.
Expensive to reverse. Pricing and packaging, data model and schema, anything customers integrate against, public commitments, and anything an autonomous agent does at scale before a human reviews it. Slow down on purpose and spend the scrutiny you used to spread across everything. Build the guardrails, the sampling, and the human review sized to the blast radius.
The point is not to slow down generally. You speed up on the large reversible majority precisely so you can afford to be deliberate on the small irreversible minority.
How to install it
Make reversibility an explicit field in how you decide. When a decision lands, the first question is not "how big is this" but "how hard is this to undo, and what is the blast radius if we are wrong." Sort into the two buckets out loud. For the agent layer, tie autonomy to reversibility: the more irreversible the action, the more human review and sampling sits on top of it. An agent drafting internal notes needs almost no oversight. An agent changing a customer-facing price needs a human in the loop, full stop.
Next time a decision lands on your desk, ask one question first: if we are wrong, can we reverse this on Tuesday? If yes, ship it and stop debating. If no, that is where your judgment goes. Spend it there.