Week 3 — Daily Async: Transisi ke Hermes

Setiap hari: 1 bacaan utama, 10-25 menit

---

📅 Senin — Workflow vs Agent: Kapan Pakai Yang Mana?

Durasi: 15 menit

Bacaan

Setelah 4 sesi n8n, kamu sudah paham workflow. Sekarang masuk ke agent.

Quick decision guide:

Langkah selalu sama? → WORKFLOW (n8n)
Input bervariasi + butuh pemahaman? → AGENT (Hermes)
Butuh reasoning/judgment? → AGENT (Hermes)
Kombinasi? → HYBRID (n8n + Hermes, Sesi 10)

Key difference:

• Workflow: kamu set rules, mesin execute
• Agent: kamu set guidelines, AI berpikir dan decide

Task

• [ ] Review use case kamu: mana yang workflow, mana yang agent?
• [ ] Identify: ada task yang lebih cocok jadi agent?

---

📅 Selasa — Hermes Setup: Spawn vs Self-Host, Model Selection

Durasi: 20 menit

Bacaan

Two ways to run Hermes:

Spawn (Managed):

• Quick setup, no server needed
• Good for: testing, learning, small projects
• Trade-off: data di cloud pihak ketiga

Self-Host:

• Full control, data di server sendiri
• Good for: production, sensitive data
• Trade-off: perlu setup + maintenance

Model Selection Guide:

Task Type	Recommended Model	Why
Classification	GPT-4o-mini / Haiku	Cheap, fast, accurate enough
Draft writing	GPT-4o	Good quality, reasonable cost
Deep analysis	Claude Sonnet	Best reasoning quality
High volume simple tasks	Gemini Flash	Cheapest per token

Task

• [ ] Install Hermes (Spawn atau self-host)
• [ ] Run hermes doctor — pastikan semua ✅
• [ ] Select default model berdasarkan use case kamu

---

📅 Rabu — SOUL.md, Memory Layers, Context Engineering

Durasi: 20 menit

Bacaan

SOUL.md = personality file agent kamu. Tanpa ini, agent generic dan tidak konsisten.

Key components:

1. Identity — siapa agent, role apa
2. Personality — traits dan characteristics
3. Tone — gaya bahasa, sapaan, formality
4. Boundaries — yang TIDAK boleh dilakukan
5. Sample responses — contoh concrete

Memory Layers:

• Short-term: per session, auto-reset
• Long-term: persistent facts, saved explicitly
• Context files: knowledge base, always loaded

Context Engineering = skill baru yang kamu pelajari. Design konteks yang bikin agent effective.

Task

• [ ] Draft SOUL.md untuk use case kamu (minimal: identity + tone + 3 boundaries)
• [ ] Identify: info apa yang perlu di context files?

---

📅 Kamis — Tools: Web, File, Browser — Read-Only Automation

Durasi: 25 menit

Bacaan

Tools = kemampuan agent berinteraksi dengan dunia luar.

3 main tools:

1. Web Search — cari info di internet

   • Good for: research, fact-checking, current data
   • Cost: ~$0.01 per search

2. File Read/Write — baca dan tulis file

   • Good for: knowledge base access, report generation
   • Cost: minimal (local operation)

3. Browser — buka dan interaksi dengan website

   • Good for: scraping, form filling, deep research
   • Cost: higher (resource intensive)

Read-only first principle: Sebelum biarkan agent modify data, start dengan read-only. Observasi dulu, baru berikan akses write.

Task

• [ ] Enable minimal 1 tool di Hermes config
• [ ] Test: minta Hermes cari info tentang topik relevan
• [ ] Review: apakah hasilnya useful?

---

📅 Jumat — Cron, Scheduling, Evaluator-Optimizer Pattern

Durasi: 15 menit

Bacaan

Cron = scheduled automation. Set once, run forever.

Common patterns:

• Daily brief (jam 8 pagi)
• Weekly report (Senin jam 9)
• Competitor check (Jumat jam 10)
• Inventory alert (daily)

Evaluator-Optimizer = quality loop:

Generate → Evaluate → Improve → Deliver

Agent bikin output, evaluate sendiri, improve kalau kurang bagus, baru deliver. Ini bikin output quality jauh lebih tinggi.

Task

• [ ] Design 1 cron job untuk use case kamu
• [ ] Write evaluation criteria untuk agent output kamu
• [ ] Preview: prepare untuk mini project Sesi 9