Human-in-the-Loop AI: Bots Prepare, Humans Decide

The question people ask about AI is whether it replaces you. That is the wrong question. The real question is whether you can review what it prepares. A tool that drafts, sources, and tracks for you, and stops the moment a prepared action would cross to the outside world, does not replace you. It gives you more room to decide.

What we set out to change

Two extremes dominate how companies use AI today. In the first, nothing is automated, so a person does every draft, every search, every follow-up by hand. In the second, an agent acts on its own with nobody able to see what it did or why. The first wastes people on toil; the second asks people to trust a black box.

The stance / how Raisy fixes it

Raisy separates the work an AI does well from the decision only a person should make. A bot prepares: it drafts the email, researches the investor, screens the candidate, tracks the follow-up. A human decides: approve, edit, or reject: at the moment the prepared work would cross a boundary. That moment is a gate, and the gate is the point.

Autonomy is earned, not assumed. A bot starts in MANUAL mode, where every prepared action waits for you. As its confidence grows from your approvals, you promote it through SEMI_AUTO to FULL_AUTO, where it runs and you audit anytime. (That lifecycle is a career path of its own, and the audit trail is what makes review real. Browse the blog.)

How it works

• AI prepares. Each bot drafts, researches, and tracks inside its process, using its kernels, the rules it knows.
• Humans gate. A run that would cross to the outside world pauses in review mode until a person approves, edits, or rejects.
• Trust is earned per bot. Approval mode is a per-bot setting; autonomy rises with a confidence score you control.
• Everything is logged. Every preparation, every gate decision, every promotion is an immutable, dated row.

What the user actually sees

You see a roster of bots, each with its role, its approval mode, and its confidence. A bot prepares an investor update; you read it, edit a line, approve. The bot's confidence ticks up. Later, you promote it to SEMI_AUTO; routine actions proceed, but the first outbound email still pauses for you. You never choose between "unreviewed AI" and "AI that never does anything", the review is the product.

Why this matters beyond convenience

Automation exists to create room for human judgment, not to erase it. When a machine drafts and a person decides, a young company can afford to attempt functions that used to require a hire, without giving up the ability to say no. The human is the final decision-maker, and the system is built so that stays true at scale.

See the gate for yourself, run one bot at raisy.team