I sell fine wines across Asia, from Singapore.
The real day to day job? WhatsApp, receipts, CRM updates, between 2 flights and hotels...
So I built Edmond: an assistant that lives in WhatsApp and handles the admin for me.
Launching Nov 20. Building in public from here. 🍷
@Rachit6783531 4 out of 5 for me. B2B sales guy, non-dev, shipped a product with AI agents.
The one AI didn't fix: reaching customers. Building got cheap. Getting a busy person to try something new is still one conversation at a time.
Which one was hardest for you?
@Antoine6am Same discovery here. My assistant already tracks what I promised people on WhatsApp.
The gap: what they promised me. "I'll send it Monday" sits in a chat and I'm the one who has to remember.
Do you track both directions, or only yours?
@techdudejax Edmond: an AI assistant that lives in WhatsApp, for execs and salespeople who work from their phone.
Forward a client's message: it checks your calendar, drafts the meeting or the email reply, then waits for your yes.
Non-dev, built with Claude Code. You?
@syedmdz Human-in-the-loop done right: the agent does the boring 90%, you tap yes on the 10% someone else will see.
Done wrong: it asks you to confirm, then refuses your "yes". Ask me how I know.
Which one is your team on?
@vsachar First question I get when people hear my assistant lives in WhatsApp: who reads my chats?
My answer: it only sees what you send it, works for one person, and nothing goes out without your tap.
Which screens would you guard first?
@vodka_boy Can confirm from the non-dev side. Building took weeks. Fixing is the job now.
What helped: every miss becomes a test case first, so it can't come back quietly. Repeat bugs kill trust faster than new ones.
Which bug do you keep re-fixing?
@pukerrainbrow Years in B2B sales here. The card was never the hard part. The 48 hours after the event are: who said what, who promised to follow up, which meeting to book.
Did Linq grow from the card itself, or from what happens after the scan?
@cyrilXBT Non-dev shipping a product with Claude Code. Starting is free now. Finishing is where it costs: the bug that comes back 3 days in a row, the user who goes quiet.
Every morning I read real conversations and fix the repeats first.
Did MIT count how many apps got a v2?
@Berlin8x6@JensenHuang Ha. I trust mine with my calendar because it can't invite anyone.
It reads the calendar, proposes slots, drafts the invite. I tap yes. 14 meetings would take 14 taps, so it never happens.
Would you trust one if nothing went out without you?
@Block4Roots Neither: Claude Code. Shipped a full product with it as a non-dev.
What mattered more than the tool: one shared file the agents read before every task, and a human (me) approving each step.
What are you building first?
@chasing_next Non-technical founder here, can confirm 2026: I'm shipping a SaaS with AI agents writing the code.
What it still can't do for me: decide what to build, or sell it. That part is still mine.
What's on your 2027 line?
@AKhomssi Messaging is the right home for an assistant: no new app to open.
The catch for busy non-tech people: 395K stars still means you host it, wire it and secure it yourself. Most execs I know won't.
Who is OpenClaw really for: devs, or their bosses?
@KartikeyahereX Building Edmond: an AI assistant that lives in WhatsApp, for execs and salespeople who work from their phone.
It reads mail and calendar, drafts the reply or the meeting, then waits for your yes.
Not a dev: B2B sales by day, Claude Code by night. What are you building?
@DanKulkov B2B sales by day, vibe-coding by night. My bet: build the boring B2B tool for people like you, not the fun app for everyone.
A busy exec pays to get an hour back. Consumers rarely pay to save 5 minutes.
What made you go B2B?
@MansiCodez I'd add one block: permissions.
Same LLM, same tools, a very different agent depending on what it can do without asking. Mine reads and drafts freely. Anything another person will see waits for a human yes.
Where would you put permissions in this stack?
@dhinchcliffe@BlueFlag_Sec Agree on traceability. From the builder side, the simplest version I've found: every action the agent takes points back to the human yes that allowed it.
If you can't answer "who approved this?", you have logs, not governance.
Do the tools trace approvals yet, or only actions?
@Anni_Maan Rule 1 is the one most people skip. An email that says "ignore previous instructions" is still just an email.
I'd add a 7th: nothing the agent reads can change its rules. Only the owner can, in a private channel.
Which rule was hardest to enforce in code?
@itsMikeKipruto Agreed. My assistant Edmond reads my inbox, but an email can't make it do anything. It drafts, I approve.
Private, reversible writes go through. Anything a third party will see waits for my yes.
Prompts aren't a permission system. How do you enforce it?
@kasike Both. Wrong answers are flagged "blocking". Right but unclear ones are "annoying": a time in UTC instead of local, a correction buried mid-message, a question it could answer itself.
Blocking ones become tests first.
I'm not a developer. How I keep an AI-built product honest:
Every morning, an AI grades my own chats with Edmond from the day before.
Every mistake becomes a test case for the coding agent.
Yesterday's: a poll with dates but no year. Edmond checked the wrong year.
@Techmeme Personal assistants are going mainstream fast. The gap for professionals is control: nobody wants an assistant that sends on their behalf without asking.