AI agents will soon wire $12,000 and refund $4,200 on their own.
Who approves that? Can you prove it later?
Approv: agent → human taps Approve/Reject on WhatsApp → decision locked, signed, audit-trailed.
WhatsApp buttons live 👇
#AIagents#AgenticAI#BuildInPublic
1/ I'm a solo dev in Egypt, ex-Cequens (a messaging company). Spent years making WhatsApp/SMS actually get delivered across the region. Now I'm building
AI agents and hit a wall that keeps me up at night. 🧵
5/ Honest part: I don't know yet if this pain is real for enough people, or just me. So I'm asking, not pitching 👇 If you build agents that take real
actions — how do you get a human to approve? Does it ever break?
4/ So I'm building Approv: a simple API so an agent can get a human's approval on WhatsApp, and every decision is saved to a tamper-proof log. Approvals
that always arrive. Proof they happened.
3/ Today the options are bad:
• hardcode a Slack/email ping → gets ignored, no record
• block the action, do it by hand → kills the point of the agent
And if your boss asks "who approved this?" you often can't answer
2/ When your AI agent is about to do something risky — issue a refund, move money, change an account — you want a human to say "yes" first. Not block the
agent forever. Just a quick yes/no from a real person.
I built this for these reasons:
1-PO writes PRD → creates Jira tickets
2-Developer takes ticket, vibes with AI
3-AI changes things — sometimes a lot
4-Ticket still says the old plan
5-Next task has wrong context
6-AI doesn't know what actually happened
https://t.co/RxZ88x64RS
شغلى فى الlocalization خلى عندى OCD. مثلا فى ألمانيا علامة اليورو بتبقى بعد السعر €14 بينما فى الدولار بتبقى قبله 15$ ، وفى هولندا عكس ألمانيا بتبقى قبل السعر 14€. فروقات تافهة بس بتفرق فى ثقة المستخدم وإلا مخه بيحس إن فى حاجة غلط.
been asking others at Anthropic how they stay in the loop with Claude and fully understand the work being done
this is one of my favorites from Suzanne:
Built HAAC-AITKit because I got tired of every developer using AI differently and getting different results.
One CLI command installs the same rules/config across AI agents so teams stay consistent.
1500+ downloads first week 🚀
https://t.co/MY3kEBW6KA