Before-and-after workflow
From fragmented effort to one guided, reviewable sequence.
Search→Rebuild context→Act manually
Organize→Guide next step→Human confirms
Fabian Delven
AI & design workflow
AI Email & Docs Specialist
An AI-assisted workspace that turns communication and documents into organized, actionable work — without taking control away from the person doing it.
Role
Product & interaction design
Type
AI workflow concept
Focus
Knowledge work · clarity
Context
Knowledge workers spend their days reading messages, finding documents, summarizing, drafting, and tracking follow-ups.
This project began as a study of that daily load. The goal was never to automate communication completely — it was to reduce the effort required to understand what actually needs attention.
The problem
Everything competes for attention at once.
Important requests are buried among low-priority messages.
Deadlines hide inside long conversation threads.
Documents are difficult to locate when they are needed.
Responsibilities are not clearly assigned.
People summarize the same information over and over.
Follow-up actions are easy to forget.
User & operational needs
What the work actually requires.
Workflow before the solution
A loop of scanning and switching.
Scan inbox
Read everything to decide what matters.
Hunt for files
Switch apps to find the related document.
Re-summarize
Rebuild context that was already understood.
Draft & track
Write a reply, then try to remember the follow-up.
Every step lives in a different tool, and the context resets each time. The effort goes into re-orienting, not deciding.
Design principles
Clarity over feature quantity.
Attention first
Show what needs a decision before anything else.
Context in place
Bring the thread and files to the decision, not the reverse.
Always a proposal
AI suggests; the person decides. Nothing is automatic.
Remove before adding
Every element earns its place or it is cut.
Proposed system
A single, calm sequence.
The experience organizes work into five steps that always end with the person in control.
- 01
Identify what needs attention
Surface the messages and documents that genuinely require a decision.
- 02
Explain why it matters
Give the reasoning — the deadline, the request, the stakes — in one line.
- 03
Connect the relevant messages and documents
Pull the related thread and files into one place, automatically.
- 04
Suggest the next action
Propose a reply, a task, or a follow-up — never assume it is final.
- 05
Let the user review and confirm
The person stays in control of everything before it is sent or finalized.
Role of AI
Assist the thinking, not replace it.
AI handles the repetitive reading and organizing so the person can spend attention on judgment.
Human confirmation points
Nothing leaves without a yes.
Three explicit checkpoints keep the person in command of the outcome:
Key screens & diagrams
One screen: attention, context, action.
Representative interface — click "Review & confirm" to try the confirmation step.
Outcome
AI can support knowledge work without removing user control.
Email and documents stop being passive information and become a clearer system for decisions and action.
5
step review-and-confirm loop
1
screen for attention, context & action
0
actions sent without confirmation
What I learned
Trust is a design material.
- Users accept AI faster when every suggestion is clearly a proposal, not a decision.
- Showing why something needs attention matters more than showing more of it.
- Fewer, well-placed confirmation points beat a wall of settings and toggles.
Next iteration
Where this goes next.
- Test the attention ranking against real inboxes to measure trust and accuracy.
- Add lightweight teammate assignment so responsibilities are explicit.
- Explore a bilingual mode so summaries and drafts adapt to the reader's language.