· 5 min read
Why every AI action in Realy waits for your approval
Autonomy is earned. We designed Realy so that AI can do almost all of the work and none of the deciding.
The fastest way to lose trust in an AI system is to let it do something you didn't expect. So we started from the opposite end: what if the AI could prepare everything, but change nothing on its own?
Reading is free, changing is a proposal
Every tool an AI employee can use is labelled. Tools that only read — searching documents, looking at the pipeline, summarising numbers — run on their own. Tools that change data create a proposal in Approvals instead, with a one-line summary of exactly what would happen.
External actions are always high risk
Sending an email, publishing a post or moving money leaves your company. Those tools are marked high risk, and an AI employee can never report one as done unless the provider actually accepted it.
Every decision leaves a trail
Each decision is audited and stays attached to its task, so you can always see what was proposed, who decided and why. Facts you keep in company memory are part of every future task — so the team gets more useful over time without ever acting behind your back.