Drafts in your style
Writes emails, memos, and documents the way you would, from a corpus of your past work: ready to send after a glance.
Executive tier · Assistant
An AI assistant that knows how you work.
Not a generic chatbot, an assistant trained on one person's context: your documents, your processes, the way you make decisions. It drafts, prepares, researches, and follows up the way a great chief of staff would, in your style.
What does an AI business assistant do?
A business assistant is a role-specific agent trained on one person's world, their documents, processes, recurring decisions, and voice. It drafts emails and documents in your style, prepares for meetings (briefs, context, talking points), researches questions you delegate, tracks your follow-ups, and acts as a knowledgeable first pass on whatever lands on your desk. Executive tier, built on Claude with your private knowledge base. Best for founders, partners, and department heads who need leverage on judgement-heavy, context-rich work that off-the-shelf tools can't template.
theagency47 · Updated June 2026 The spec sheet
How the Business Assistant actually works.
| Job-to-be-done | Act as a context-rich assistant for one role: draft, prepare, research, and follow up in that person's style |
| Tier | Executive (decision support) |
| Underlying model | Anthropic Claude (Opus/Sonnet by task) |
| Trigger | On-demand requests + standing routines (daily prep, follow-up sweeps) |
| Inputs | Private knowledge base (docs, SOPs, past decisions), email/calendar access, voice/style guide |
| Tools available | Email/calendar read+draft, docs, web research, CRM/PM read, task tracker |
| Autonomous decisions | Draft framing, meeting-prep scope, research depth, follow-up prioritisation |
| Escalation rules | Anything to be sent/committed externally → human approval · Sensitive decision → presents options, doesn't decide |
| KPIs measured | Tasks handled, draft acceptance rate, prep time saved, follow-ups not dropped |
| Eval suite | 20 test cases (style match, context recall, appropriate escalation). |
What it does
The work it takes off your team.
Meeting prep
Before each meeting, assembles who's attending, the history, the open threads, and suggested talking points.
Delegated research
Takes a question you'd normally hand an assistant and comes back with a grounded, sourced answer.
Follow-up tracking
Remembers what you committed to and who owes you what, and nudges before things slip.
Knows your context
Trained on your SOPs, decisions, and documents: so answers fit how your business actually runs.
First-pass triage
Acts as an informed first pass on inbound asks, surfacing what truly needs you and handling the rest.
Integrations
Standard integration stack.
- Email / calendar: Google Workspace, Microsoft 365
- Knowledge: private KB from your docs / Drive / Notion
- Docs: Google Docs, Notion, Office
- CRM / PM: read access for context
- Tasks: your task tracker for follow-ups
Custom integrations with proprietary systems are quoted as add-ons. Not sure if yours fits? Describe your stack and we'll confirm.
What it looks like in practice
A typical run.
- Early, Daily brief: today's meetings prepped, priorities surfaced, anything time-sensitive flagged.
- On request ("Draft a reply to this," "research X," "summarise this thread") handled in your style.
- Across the day, Tracks commitments and follow-ups, nudging before deadlines slip.
- Always, Anything outbound waits for your approval; sensitive calls are presented as options, not decided.
FAQ
Questions about the Business Assistant.
How is this different from ChatGPT?
Generic chat has no durable memory of your business. This assistant is built on your private knowledge base and trained on your style, so it answers with your context and sounds like you. For team-wide enablement on frontier models, see ChatGPT / Claude for Business.
Is my data private?
Yes. Your knowledge base is yours; we use enterprise terms with no training on your data, scoped access, and a DPA before deployment. It's built around your privacy from the start.
Will it act without me?
Not on anything that leaves the building. Outbound messages and commitments wait for your approval; on sensitive decisions it presents options rather than deciding. You set how much autonomy it has.
How is it delivered?
As a Spark build, or as the executive-tier agent in a Workforce Pro engagement.
Related