On the Idea Wall · ▲ 0 votes
FlowDesk — the spec
FlowDesk is an AI agent that watches inbound email, extracts actionable items (meetings, deadlines, follow-ups), and updates a unified calendar and task board automatically. It is built for small management teams whose clients are weak at self-managing — teams that currently lose time chasing confirmations, manually copying dates, and tracking commitments across scattered tools.
Decisions locked
| Question | Answer |
|---|---|
| Shape | AI agent / automation — acts on data without prompting the user each time |
| Audience | Small management teams handling disorganised, high-maintenance clients |
| Core loop | Email arrives → AI extracts event or deadline → calendar and task board update automatically |
| Success metric | After 30 days, the team runs a full client workflow end-to-end inside FlowDesk with no external tool required |
The core loop
- Email or SMS arrives from a client or external party into a connected inbox.
- AI agent parses the message, identifying meeting requests, deadlines, action items, and the relevant client record.
- Calendar updates automatically — new events are created, existing ones rescheduled, and conflicts flagged.
- Task board reflects new actions — deadlines and follow-ups are added with due dates and assigned owners.
- CRM record is updated — the relevant client timeline logs the interaction and any commitments extracted.
- Team is notified only when a human decision is actually required (conflict, ambiguity, or high-priority flag).
v1 scope
Email Parsing Agent — monitors connected inboxes in real time, reads plain text and structured content, and extracts meetings, deadlines, and tasks using an LLM. Lives in the background; surfaces output on the Dashboard.
Auto-Calendar Management — creates, updates, and cancels calendar events from extracted data without manual entry. Flags conflicts for human review on the Calendar screen.
Unified Task Board — converts extracted action items and deadlines into tasks with due dates, priorities, and assignees. Lives on the Tasks screen; updates in real time as new emails are parsed.
Client CRM Log — maintains a per-client timeline of all parsed interactions, commitments made, and status changes. Lives on the Client Profile screen; imported from spreadsheet on first setup.
Conflict and Ambiguity Alerts — when the agent cannot resolve a scheduling conflict or extract intent with high confidence, it surfaces a single actionable prompt to the relevant team member. Lives in a notification tray on the Dashboard.
Deliberately later
- SMS parsing — connectivity with SMS gateways adds carrier integration complexity; stabilise email parsing quality first before expanding input channels.
- Outbound AI replies — auto-responding on behalf of the team is a trust and liability question; earn that trust after the team validates the agent's extraction accuracy over 30 days.
- Client-facing portal — giving clients direct access shifts the product's surface area significantly; address once internal workflow is proven.
- Advanced analytics and reporting — useful after the team has enough data volume (post-30-day threshold) to make reporting meaningful.
- Workflow automation builder — a no-code rule editor belongs in v2 once the team knows which recurring patterns are worth automating.
Data model sketch
client— id, name, contact details, assigned team member, status, created_atemail_message— id, client_id, raw_body, source_address, received_at, parsed_statusparsed_extraction— id, email_message_id, extraction_type (meeting / deadline / task / ambiguous), confidence_score, resolved_bycalendar_event— id, client_id, extraction_id, title, start_time, end_time, conflict_flag, created_attask— id, client_id, extraction_id, description, due_date, assignee_id, priority, statuscrm_log_entry— id, client_id, source_type (email / manual), summary, logged_atalert— id, extraction_id, alert_type, resolved, resolved_by, created_atteam_member— id, name, email, role, connected_calendar_token
Screens
- Dashboard — live feed of today's extractions, pending alerts, and team activity at a glance.
- Calendar — unified calendar view showing all auto-created and manually added events, with conflict flags.
- Tasks — kanban or list board of all extracted action items grouped by client and due date.
- Client Profile — individual client record showing CRM log, linked events, open tasks, and contact history.
- Client List — searchable directory of all clients with status indicators and quick-access links.
- Inbox Monitor — configuration screen for connected email accounts, parsing rules, and confidence thresholds.
- Alerts Tray — queue of unresolved ambiguities and conflicts requiring a team member decision.
How to use this document
- Bring it into Mind Palace, where the guided platform and its AI coach pick up from exactly this document and build the product with you step by step.
- Hand it to a developer as the complete v1 brief — scope, data model, and screens are defined and ready.
- Keep it as the single source of truth for every team and stakeholder conversation while you build, so scope stays locked.
This is Esther Sande's. Build yours.
A spec like this is step 3 of Mind Palace's nine — the other six take a weekend: working software, a real offer, first customers.
Want to build this with Esther Sande?
Tell us who you are and we'll broker the introduction — their contact details stay private until they say yes.