What we built, how it works, and what we're inviting people into.
What this session is
An intimate, invitation-only session hosted by SORRYWECAN — built for strategic partners and select Mind Shift alumni who are ready to move from curiosity to application. Two hours of practical intelligence: the latest AI models and tools, how we use them daily, what they actually cost, and how to wire them into your business or creative process. No theory. No filler. Just what works.
Free for invitees~2 hoursCurated, closedPractical-firstMind Shift alumni + partners
What we built these past days
🧠 Autonomous AI Agent Stack
A self-running intelligence system — The Master — that wakes up every morning, reads the internet, filters what matters, and delivers a curated brief to Telegram before 8 AM. No manual steps. Fully automated.
⚙️ Agent Harness
Every script runs inside a fault-tolerant wrapper: 3 retries on failure, timeout protection, Telegram alerts if something breaks, and a cost tracker that logs every token spent. Weekly spend report delivered every Monday at 10 AM.
RSS curation with content-hash dedup (never shows the same article twice) → Telegram
🔍
Midday + Evening Signal Scans
X/Twitter + signal pipeline → only fires if 4+ strong signals detected
🌐 research-hub.xyz
Live research site with 19+ blog posts, an auto-updating /intelligence section (Slovak), and a /radar page. Deployed on Vercel, content pushed from the agent automatically.
Model stack (live and authenticated)
Claude Sonnet 4.6
Primary — orchestration, briefs, writing
Claude Haiku 4.5
Worker — fast scripts, data pipelines
Gemini 2.0 Flash
Fallback — speed, multimodal
DeepSeek V3
Fallback — cost-efficient reasoning
DeepSeek R1
Deep research — long-form analysis
Claude Sonnet (OpenRouter)
Redundancy fallback
Session curriculum (draft)
01 — The landscape right now (15 min)
What models exist, what they're actually good at, and what the real cost of using them is. Claude vs Gemini vs DeepSeek — practical comparison, not hype.
02 — How we use it daily (30 min)
Live walkthrough: morning brief, signal scanning, automated research. What's running, how it's built, what it saves us in hours per week.
03 — Apply it to your business (45 min)
Workshop format. Participants map their own workflows. Where does AI replace a recurring task? Where does it assist? Where does it still fail? Honest answers.
04 — Setup & tools (20 min)
What to actually install. What APIs are worth paying for. How to start without being a developer. The minimum viable AI stack for a small team or solo operator.
05 — Open Q&A (10 min)
Specific questions, edge cases, and what's coming next.
How The Master works (for internal context)
🔄 Memory system
Daily logs in memory/YYYY-MM-DD.md capture what happened each day. Long-term curated memory lives in MEMORY.md — distilled every few days, like a human reviewing their journal. The agent updates its own memory without being asked.
🎯 Smart routing
Tasks are classified and routed to the right model automatically — cheap Haiku for scripts and data, Sonnet for synthesis and writing, R1 for deep research. Cost-aware by design.
🛡️ Resilience
3 retries on every failure. Timeout protection per job. If something breaks 3 times in a row, it logs the rule to AGENTS.md so the system learns and doesn't repeat the mistake.