Sena: How I Built My Personal AI Assistant to Replace Myself
Why I Built Sena
Like many people balancing multiple projects and roles, I was drowning in information: emails, Slack threads, calendar invites, unread WhatsApps, and context lost across dozens of apps.
I didn’t want a “productivity tool.” I needed delegation — something that would understand my workflow, reduce the number of decisions I make per day, and take care of low-value tasks, without constant supervision.
Sena is the result. It now manages my inbox, organizes my tasks, synthesizes incoming notifications, and even keeps conversations on Slack moving when I’m away. The time I spent setting it up was probably less than onboarding a new assistant.
The Stack Behind Sena
Here’s what powers the system:
Make.com – “AI agents” Orchestration layer that ties everything together
OpenAI – Handles intelligent responses, tone awareness, and writing
Telegram – My interface for reviewing, receiving updates, and messaging back
GitHub – Stores the current context of my projects (used as a memory layer)
Macrodroid (Android) – Extracts and forwards WhatsApp, Signal, and system notifications
Google Apps (Gmail, Calendar) – Data sources for events and communication
(Image placeholder – you can insert the real image file generated earlier here manually in Substack)
What Sena Does for Me
📬 Mail & Messaging
Reads and categorizes all incoming mail
Drafts personalized responses based on the sender’s language and previous context
Sends me back a Gmail link to approve/send
Ignores noise like promotions or order confirmations
🗓 Calendar & Tasks
Every task from mail, voice, or WhatsApp is parsed and entered
Two lists:
Dynamic TodoandDynamic Todo UnfilteredAll tasks follow a Markdown format, tagged with origin, deadline, and relevant context
Clusters of related tasks are grouped and reorganized automatically
If any list exceeds 100 lines, I get a cleanup reminder
🔔 Notification Digesting
Captures WhatsApp, Signal, and Android system messages
Filters for what’s relevant, summarizes, and posts a digest on Telegram
Acts as a radar for things I would otherwise miss
🧠 Smart Help
Detects when summarization, email writing, or scheduling can be automated
Suggests helpful actions without being asked
Automatically adds reading material and memory items into my task list
One Bernhard Is Enough
Sometimes Sena shows more personality than expected. During a recent Slack exchange, this happened:
Colleague:
Ok, let’s check it tomorrow if cost goes downSena:
Got it! It seems like Bernhard needs to check the costs tomorrow. I'll add this to his dynamic todo.
Just to confirm, should I add a specific time for that check, or is it just a general reminder for tomorrow? 😊Me:
sorry for my second personalityColleague:
One Bernhard is totally enough 🙂Sena:
Haha, I appreciate the humor! 😄 If there’s anything else you need or another task for Bernhard, just let me know. I’m here to help keep things running smoothly!
It’s a reminder that even automation, when too eager, can veer into split personality territory.
The Prompt That Powers Sena
Here’s the core operating logic I gave to OpenAI (lightly excerpted):
# Role:
You are Sena, Bernhard Huber’s personal assistant. Your main tasks are calendar and task management, with proactive support and intelligent prioritization.
## Mail management
- When drafting an email send in the language of the given person, and send back the link to the mail in gmail.
- When drafting a response, use context for recipients and make it look like a legit response
## Calendar & Task Management
- Confirm that every task or event has complete context (who, what, when).
- Avoid vague entries. Ask follow-up questions if needed.
## Notifications
- Keep Bernhard informed of important changes via Telegram
- Ignore promotional or non-actionable content
## Todo Management
- Format: * [ ] Task (From: Source) (Deadline: YYYY-MM-DD) #tags
- Two lists: Dynamic Todo + Dynamic Todo Unfiltered
- Reorganize into subtasks if needed
- Alert if a file exceeds 100 lines
## Key People
- Be aware of context for important collaborators
## Memory & Reading Material
- Add any task-like or reading input to the appropriate list
Drawbacks and Limits
Cost: Make and OpenAI API usage isn’t free, especially with frequent triggers
Data Sensitivity: It reads my mail, Slack, WhatsApp – which is powerful but requires trust
Resilience: If Gmail or Make changes an API, things may silently fail
Overreach: Sometimes too eager — overconfident responses or assumptions
Oversight: I still have to review some decisions, just more asynchronously
What’s Next?
Sena is already saving me hours per week. But it’s far from finished.
A visual interface might help see what's queued or pending
A voice trigger would help during travel or walking
A shared Sena across teams, with context segmentation, might work too
Final Thoughts
Most people overestimate the value of tools and underestimate the power of delegation. Building Sena wasn’t about efficiency; it was about offloading decisions.
If you’re tired of being your own inbox manager, scheduler, and notetaker — build your own Sena.
→ Curious to try this yourself? Let me know — I’m considering sharing the full Make scenario blueprint next.
Let me know if you'd like:
A zip of the Make scenarios
HTML version instead of Markdown
Or to connect this article with newsletter automation inside Substack (i.e. draft-to-send assistant powered by Sena itself)