TL;DR
- Three AI agents running on one Telegram account: e-commerce automation, Japanese tutor, priority scheduler
- E-commerce bot: Instagram research → product renders → Shopify pages → Meta Ads (72-hour build)
- Japanese agent: 10 words/day with enforced written and spoken usage
- Best for: Entrepreneurs automating business + personal development
- Key lesson: Telegram as universal control plane—one interface for multiple agents
Shaun Smerling built three distinct AI employees that all report through the same Telegram account.
When people ask “what has anyone actually built with agents,” Shaun has an answer: a full e-commerce pipeline, a language tutor that refuses to let you skip practice, and a scheduling system that realigns your priorities monthly. All managed from his phone.
The E-Commerce Bot (72 Hours)
The first agent runs a complete product pipeline. It starts on Instagram, scanning design trends and concepts. When it finds something promising, it sends the idea to Shaun via Telegram for approval.
Approve it, and the agent takes over: generates product renders using Nano Banana, writes Shopify product descriptions, creates the listing, and launches Meta Ads campaigns.
Build time: 72 hours from scratch.
Tech stack: Telegram Bot API, Instagram integration, Shopify API, Meta Ads API, Nano Banana for renders.
This isn’t a demo. It’s running a real e-commerce operation—research, product creation, and ad launch—while Shaun focuses on strategy and approval gates.
The Japanese Tutor (Built in Days)
The second agent teaches Japanese. But it’s not flashcards.
Every day: 10 new words. The agent introduces them via Telegram, then quizzes Shaun. But here’s the enforcement layer: it doesn’t accept passive recognition. You must use each word in a coherent sentence—both written and spoken.
No shortcuts. No “I know this one, skip it.” Active production or the word doesn’t count.
Build time: A few days.
Why it works: Most language apps test recognition (do you remember this word?). This agent tests production (can you actually use it?). Recognition is passive. Production builds fluency.
Shaun checks Telegram in the morning. The tutor is waiting. No app to open, no separate routine to remember.
The Priority Scheduler (Work in Progress)
The third agent is still being refined. It takes Shaun’s monthly prioritization input and continuously realigns his schedule to keep him moving toward the highest-value work.
Not just “remind me at 3pm.” More like: “You said monthly revenue growth was priority one. Your calendar shows 12 hours this week on low-priority admin. Reallocate?”
It’s less about task management, more about strategic alignment over time.
Status: In progress. Shaun was “tinkering with it today” as of March 1.
The Pattern: Telegram as Control Plane
All three agents interface through Telegram. One account. One notification stream. One place to approve, reject, respond, and review.
You’re not logging into three dashboards. You’re not switching contexts. The agents come to you—on the device you already check 50 times a day.
This is the shift: AI as employees, not tools. Employees report through a unified interface. Tools scatter across separate apps.
Shaun built these agents to answer skeptics. “What has anyone actually built with agents?” Here’s the answer: systems that run businesses, teach languages, and realign priorities—all managed from a phone.
