Demo · 1 min read
Automating your company, part 7
Yes/no approvals based on facts
Hint: watch the video at 2x speed
Predictive analysis and AI based on facts look like opposites. In the end they touch. To predict is to read past behaviour and extrapolate it into variations of the same thing, and agents already carry the mathematical and computational models to do that. So there is not much left to do but ask them: predict something, or do something based on facts.
Here we ask for facts. One prompt points the Orchestrator at a brain holding a company's invoices, messages and a notes.md with its rules:
/approvals for @agent:orchestrator using @brain:brain_0e19fd8d-ae5 — propose at most 5 next tasks based only on the facts in notes.md, each with its evidence and the notes.md rule it follows; Yes sends the task as drafts only (never send, never delete), No skips it, and add a Why? button that asks the brain to explain the item.
It comes back with a small page of proposed tasks. Mar Azul, a client since 2024, can pay in two instalments. Lumen Café, new this year, cannot. Each task shows its evidence and its rule. Yes runs it as a draft, No skips it, and Why? asks the brain to explain.
Your job becomes a few yes or no answers.
Run your company with a yes or no
Join the waitlistDemonstração · 1 min de leitura
Automatizar a sua empresa, parte 7
Aprovações sim/não baseadas em factos
Dica: veja o vídeo em velocidade 2x
A análise preditiva e a IA baseada em factos parecem opostas. No fim, tocam-se. Prever é ler o comportamento passado e extrapolá-lo para variações do mesmo, e os agentes já trazem os modelos matemáticos e computacionais para o fazer. Por isso resta pouco mais do que pedir-lhes: prevê algo, ou faz algo com base em factos.
Aqui pedimos factos. Um único prompt aponta o Orchestrator a um cérebro com as faturas, as mensagens e um notes.md com as regras de uma empresa:
/approvals for @agent:orchestrator using @brain:brain_0e19fd8d-ae5 — propose at most 5 next tasks based only on the facts in notes.md, each with its evidence and the notes.md rule it follows; Yes sends the task as drafts only (never send, never delete), No skips it, and add a Why? button that asks the brain to explain the item.
Ele devolve uma pequena página de tarefas propostas. A Mar Azul, cliente desde 2024, pode pagar em duas prestações. O Lumen Café, cliente novo este ano, não pode. Cada tarefa mostra a evidência e a regra. Sim executa-a como rascunho, Não ignora-a, e Porquê? pede ao cérebro que explique.
O seu trabalho passa a ser algumas respostas de sim ou não.