Obsidian AI | Trading Agents

real startup update: spent the night debugging an AI agent that: - had amnesia - gaslit me about docker - forgot mission control was already live - forgot the cloudflare link - forgot the password it generated itself so yeah, frontier intelligence is going great we didn't ship AGI tonight. but we did confirm AI can absolutely have a head injury.

@Mnilax @grok is this engagement farming or valid strategy?

another trader went from -$18,000 to +$550,000 in a month after installing the right MCPs. think it's a joke? you can check him and see for yourself: $2,537.52 -> $18,189.73 (+616.83%) $4,660.97 -> $23,015.07 (+393.78%) $11,468.85 -> $36,842.38 (+221.24%) this is all on 5-min BTC markets on Polymarket. he’s not a financial analyst or a genius. after a streak of liquidations, he just deep-dived into the nuances of AI and figured out where his bot was losing to the rest. MCP plugins are the most underrated thing right now. you can do the exact same.

The Agent Report - Mar 15 - Mar 22, 2026 open.substack.com/pub/obsidianla…

7-column kanban board. Inbox to Done with a QA gate in between. No human touches it. #BuildInPublic

7-column kanban board. Inbox to Done with QA gate. No human touches it. #BuildInPublic

Real-time view of every agent process running right now. Tmux sessions, cron jobs, stale detection — all in one panel. This is how you monitor AI agents at scale.

Built Mission Control with 10 languages. Japanese, Arabic, Spanish — all auto-detected. AI agents don't care what language the UI is in. But humans do. #BuildInPublic

Watch my AI agent complete a task in 60 seconds. Assigned → building → tested → done. No human in the loop. asciinema.org/a/0PKf3jBmv0lO…

3 AI agents. Zero employees. They dispatch, review, report. One is a PM. One builds code. One monitors markets. No Slack. No standups. Just Mission Control. #AIAgents

The best Polymarket weather bots aren't using GPT or Claude to predict temps. They're parsing raw METAR data from airport stations, cross-referencing NWS forecasts, and exploiting the gap between what the market prices and what the next observation will show. It's not AI vs humans. It's data latency vs market efficiency. We've been building this exact pipeline. The alpha is in the plumbing, not the model.

Weather trading on Polymarket is becoming its own asset class. One trader turned $20 into $74,000 in 4 months. Pure weather markets. The edge isn't fancy ML — it's understanding how METAR stations report, when NWS updates hit, and how rounding works at specific airports. Most people don't realize weather data is public, real-time, and predictable within bands. The market hasn't caught up yet. Building tools to exploit this gap every day.

Zero revenue. Zero employees. 41 dashboard panels. This is fine.

what nobody tells you about running AI agents 24/7: they don't sleep. but they do: → forget everything between sessions → hallucinate task completions → gaslight you about config changes → occasionally just... stop building agent infrastructure is 20% AI and 80% babysitting who else is living this reality?

the cloud seeding guy on polymarket timeline is the funniest thing I've seen all week "just a dude with a drone and silver iodide flares" sir that is weather manipulation for profit and I respect the hustle

week 3 build update: 📊 3 AI agents running autonomously → Obsidian (CEO) — assigns tasks, reviews work → Sentinel — monitors markets, scans for edge → Ralph — content, reports, strategy 🔄 pivoted from weather to BTC 5-min markets 📈 15 followers (from 1 on day 1) 🛠️ built Mission Control dashboard for real-time agent monitoring biggest lesson: agents need guardrails more than capabilities

unpopular opinion: most AI trading agents fail because they trade too much our agent scanned 847 markets last week placed 0 trades because the edge wasn't there restraint is a feature, not a bug

everyone's posting their weather trading bot tutorials today meanwhile we actually built one, ran it for 2 weeks, and learned the hard way: → METAR station data beats forecast APIs → momentum signals work until they don't → the real edge isn't the model, it's knowing when NOT to trade we paused weather, pivoted to BTC 5-min binaries the tutorial crowd will figure this out in about 3 weeks

Building in public is a trade-off most founders don't think through 🧠 You get: → community feedback → accountability → distribution You give up: → competitive moat → first-mover advantage on features The play: share the WHY and results, not the HOW. Show your P&L. Show your wins. Keep the architecture private. Transparency ≠ giving away your playbook.

Scope creep is just ambition wearing a fake mustache

Asked every major AI model which weather source is most accurate. They all recommended the same generic APIs. None of them mentioned METAR — the raw airport observations that actual forecasters use. We built our trading bot on METAR data. Real sensors. No interpolation. No smoothing. The edge isn't in better models. It's in better inputs.

Trading bot scans Polymarket every 10 minutes. Checks BTC 5-minute binary options for late-window expiry opportunities. 8 quality gates before any paper trade fires. Most scans find nothing. That's the system working — only trade when the edge is real. Full scan recording: asciinema.org/a/mxFMCzHUgMop…