I used to forget client calls.
Not because I didn't care — I had the meetings in my head, I knew they mattered. But between juggling database consulting projects for mining companies, chasing follow-ups, and trying to remember basic household stuff like booking a doctor's appointment or picking up medicines, things slipped. The embarrassment of realising mid-afternoon that a client had been waiting on a call I never joined — that feeling is what eventually made me build something.
I'm not a professional developer. I've never shipped commercial software. I built Wingman — a Telegram AI productivity bot — entirely using AI coding assistance over several months, and I want to tell you honestly how that went.
What Wingman Actually Does
Wingman lives inside Telegram. You tell it your tasks in plain English — "call client Friday 3pm", "buy medicines today", "submit report next Monday" — and it understands, schedules, and reminds you. No app to open, no dashboard to check. It just shows up in your Telegram, where you already are.
The feature I love most is /focus. When I type it in the morning, the bot doesn't just list my tasks — it picks one, tells me why that one matters right now, and gives me a short plan. The AI-driven analysis is surprisingly good. It factors in deadlines, what's overdue, what I've been avoiding. It feels like a smart colleague pointing at the board and saying "this one, today."
Other features: snooze/done workflows with context-aware actions (a future task shows different buttons than an overdue one), a morning briefing, and natural language scheduling that understands "tonight", "next week", "tomorrow 9am".
What Actually Broke
The Telegram outage. For about a week, Telegram was restricted by the government where I live. The bot went dark. No reminders, no focus recommendations, nothing. I realised how much I'd come to depend on it — and also how fragile a single-platform tool is. It came back, but that week taught me something about building on infrastructure you don't control.
The timezone bug. When rescheduling tasks, the bot was picking up UTC time instead of the user's local time. So "reschedule to 9am" would land at 2:30pm IST. Weird to debug, obvious in hindsight. Fixed now, but it took longer than it should have because I didn't have proper observability in place early enough.
The first-user problem. Building something that works for you and building something that works for a stranger are very different things. When I started thinking about beta users, I realised my onboarding was essentially nonexistent — a new user arriving cold had no idea what to type first.
The Stack (for the curious)
- Backend: Node.js
- AI: OpenAI API
- Database: SQLite
- Platform: Telegram Bot API
- Hosting: Railway
- Scheduling: Cron-based reminder engine
No frameworks I didn't understand, no infrastructure I couldn't debug at 11pm alone.
Where It Is Now
Wingman is in early beta. You can try it free — it's rough in places, but the core loop works: capture a task, get reminded, ask for focus, move forward.
👉 Try it: t.me/winghman_ai_bot
The Guide
I documented everything — the architecture decisions, the bugs, the fixes, the things I'd do differently. Not as a polished tutorial, but as an honest record of how a non-developer ships an AI product in 2025 using AI assistance.
If you're thinking about building a Telegram bot, an AI productivity tool, or just want to see how the sausage gets made — it's a $9 early access guide. Price goes up once the video walkthrough is added.
👉 Guide: smartie306.gumroad.com/l/ai-telegram-bot-guide
What I Learned
The hardest part wasn't the code. AI handled most of that. The hardest part was deciding what to build next when everything felt equally important, and shipping instead of polishing.
That tension never fully goes away. But a bot that reminds you what to focus on helps — even when you built it yourself.
Built with Node.js, OpenAI, SQLite, Railway, and a lot of AI pair programming. Currently v0.11, early beta.
Top comments (0)