Move with
confidence.
Useful autonomy starts with clear boundaries. We’re building systems whose actions can be inspected, questioned, and corrected.
See the principlesYour direction stays yours.
Permissions should stay clear, specific, and changeable.
Actions should be understandable.
Know what was attempted, what happened, and what needs attention.
Confidence needs
something to stand on.
When an agent can act, care has to show up in what it is allowed to do, what you can see, and how you can step in.
Safety is practical. It means limiting what can go wrong and giving you a clear way to respond when something does.
Choose the scope.
Help with a draft should mean help with a draft. Sending it is a separate choice. Permission needs to be specific enough that you know what you’re agreeing to.
See what happened.
An attempt and a result are different things. You should be able to tell what an agent tried, what actually happened, and what remains uncertain.
A useful account gives you enough context to check the outcome and decide what comes next.
Change your mind.
A plan can change. So can your permission. You should have a clear way to pause, narrow the task, correct a misunderstanding, or stop.
Continuing the work must stay connected to your present direction.
Keep a way back.
Some actions are easy to undo. Others affect people or systems beyond the conversation. The care around an action should reflect its consequences.
Where an action can be reversed, preserve that path. Where it cannot, make the stakes and the decision clear before proceeding.
Your direction stays yours.
That includes the right to pause, correct, or stop.
Care has to become practice.
AI can be useful and still get things wrong. As agents gain the ability to act, safety depends on the conditions around those actions: the permissions, the checks, and the response when an outcome is uncertain.
Accountability makes those conditions answerable. It asks whether the action matched the person’s intent, whether the outcome can be checked, and how a mistake can be addressed.
These are expectations we design around across Long Arc and Nellie. Any claim that a product meets them needs to be supported by its actual behavior. Confidence should grow from what people can understand and verify.
That is how accountability supports continuity. The work can continue with a clear account of where it stands, while the person remains free to change its direction.