Core concepts

Four ideas explain almost everything you’ll see in the product. Each exists to answer a different question a founder actually asks.

The operating picture

“How are things, really?” — answered once a day. Every morning Nerve reads your connected rows — revenue, pipeline, spend, operations — and assembles a briefing: what changed while you were gone, what needs you most today, and the numbers behind both. It also carries forward-looking calls (a deal likely to stall, a cost trend about to cross a line), each with a confidence level.

Predictions are tracked to resolution and graded in the open. Until enough calls have resolved to be statistically meaningful, Nerve shows the honest count — calls made, calls resolved — rather than an accuracy percentage it hasn’t earned.

Cited or abstain

“Can I trust this number?” — the rule that decides what Nerve is allowed to say. Numbers are computed from your rows in code; the AI explains what was computed, it never invents the figure. Every claim carries a citation back to its evidence, and you can open any answer to see its sources.

When evidence is missing or too thin, Nerve abstains: “I can’t answer this yet, here’s what I’d need.” Statistical claims sit behind fixed minimum-sample gates — comparisons and rankings simply don’t render until there’s enough data to make them honestly.

The debate

“Should I actually do this?” — for hard calls, one confident answer isn’t good enough. Nerve convenes multiple AI voices with different mandates over the same evidence pack: positions argue for and against, a devil’s advocate attacks the emerging consensus, and the voices vote. You get a verdict with the vote count and the dissent preserved — 4–2 is reported as 4–2, and a deadlock is reported as “no majority,” never a manufactured winner.

The whole debate is an artifact you can audit: the question, the clash, the attack, and a verdict whose claims cite evidence IDs. Agree or disagree — you’ll know exactly with which premise.

Workflows and approval

“Can it do the work without me losing control?” — Nerve’s workflows trigger on your rows, not on a calendar. When one fires — say, a deal has gone quiet past your threshold — it does the safe part of the work (drafting the follow-ups, preparing the update) and then stops at the last safe step. Nothing outbound sends without your approval.

Autonomy is earned, per workflow, from its own track record:

• —Draft & ask — the default. Nerve prepares the work and waits for you.
• —Act & notify — for workflows you’ve promoted: it acts, and tells you immediately.
• —Full auto — reserved for workflows with a proven record, granted explicitly by you.