Update dashboard, memory, root (+8 ~6)

memory/kb/youtube/2026-04-01-claude-mythos-changes-everything.md

@@ -0,0 +1,138 @@
+# Claude Mythos Changes Everything. Your AI Stack Isn't Ready.
+
+**URL:** https://youtu.be/hV5_XSEBZNg  
+**Durată:** 31:20  
+**Data salvare:** 2026-04-01  
+**Tags:** @work @growth #ai #claude #workflow #prompt-engineering
+
+---
+
+## 📋 TL;DR
+
+Claude Mythos (Capybara) - primul model antrenat pe Nvidia GB300 - va fi lansat în 1-2 luni și va schimba fundamental modul în care construim cu AI. Mesajul central: **simplificarea este cheia**. Modele mai puternice înseamnă mai puține instrucțiuni procedurale, mai puține scaffold-uri complexe, mai multă încredere în capacitatea modelului de a înțelege outcome-ul dorit. Trebuie să ne pregătim ACUM pentru această schimbare, altfel vom rămâne în urmă.
+
+---
+
+## 🎯 Puncte Cheie
+
+### Ce este Mythos?
+- Primul model antrenat pe cipuri Nvidia GB300
+- Noul lineage: **Capybara** (nu mai e Sonnet/Opus)
+- Cel mai mare și mai puternic model din lume (conform Anthropic)
+- **Exemplu impact:** A găsit zero-day vulnerabilities în Ghost (50k stars GitHub) pe care security researchers nu le-au găsit niciodată
+
+### 4 Zone de Audit Pre-Mythos
+
+#### 1️⃣ Prompt Scaffolding
+- **Întrebare cheie:** "E această instrucțiune aici pentru că modelul o NECESITĂ sau pentru că EU am vrut să o necesite?"
+- Anthropic: "Consider adding complexity only when it demonstrably improves outcomes"
+- OpenAI Codex: "Just tell it what you need without writing long instructions"
+- **Exemplu:** Prompt de customer support cu 3000 tokens - jumătate procedural → poate dispărea 30-50% când vine Mythos
+- **Regula:** Cere WHAT + WHY, nu HOW
+
+#### 2️⃣ Retrieval & Memory Architecture
+- Mai puțină logică de retrieval pe partea ta, mai multă încredere în model
+- Nu RAG e mort, dar modul de gândire se schimbă
+- **Noul pattern:** Prezintă un repo bine organizat → spune modelului să caute ce are nevoie
+- Modelul devine mai bun la a-și umple context window-ul eficient
+
+#### 3️⃣ Domain Knowledge Hardcoding
+- **Întrebare:** Ce reguli de business le-am scris pentru că modelul NU putea infera vs ce poate infera acum?
+- **Exemplu personal:** Prompt de research de 10 linii → o linie → rezultate mai bune (prompt-ul detaliat LIMITA modelul)
+- House style pentru rapoarte → poate fi inferat din exemplu, nu mai trebuie specificat
+- **Arta promptării evolueză:** Din "ce pui în prompt" → "ce lași afară"
+
+#### 4️⃣ Verification & Evals
+- Ne mutăm de la 85% correct → 99% correct
+- **Recomandare:** UN SINGUR eval gate la final care verifică TOTUL
+- Nu mai pierde timp cu intermediate evals - simplifică pipeline-ul
+- Pentru non-tech: păstrează standarde ÎNALTE - nu accepta "99% e bine" dacă acel 1% contează
+- Pentru tech: evals automate comprehensive - omul devine bottleneck la volume
+
+### Implicații Practice
+
+**Cost & Access:**
+- Modele foarte scumpe (probabil doar pe Max plan la început - $200/lună)
+- Trebuie să te gândești: investesc în "cutting edge curve" sau stau un pas în urmă?
+- **ROI:** Dacă ai acces la Mythos, profită la maxim - poate compensa $200 găsind economii în subscripții
+
+**Simplificare Generală:**
+- Modele scumpe → folosește tokens eficient
+- Nu le aglomera cu proces descris de oameni
+- Lasă modelul să decidă ordinea tool calls, ce să pună în context, etc.
+
+**Multi-Agent Patterns:**
+- Mythos devine **planner** - tu dai outcome spec + evals + tool suite
+- Mythos instanțiază agenți, măsoară progres, verifică cu evals
+- Pattern: agent separat face eval (nu același care a făcut treaba)
+
+**Non-Tech Work:**
+- "Under the desk software" devine din ce în ce mai sofisticat
+- Construiești aplicații utile fără să atingi cod - doar specificând intent
+- Exemple: family calendar, team tools, workflows complexe
+
+### Scaling Law & Bitter Lesson
+
+> "The bitter lesson we have to learn: all the way we have described process, the things that are precious to us are things that are associated with our ability to execute work in a certain series of steps and somehow we've decided that's an important reflection of our work identity. What Claude Mythos and similar models are going to teach us is that **that doesn't matter anymore** and what matters is **the outcome** and our ability to **name the outcome and let go of the process**."
+
+- Simplitatea funcționează mai bine pe măsură ce modelele devin mai puternice
+- Contribuțiile noastre "speciale" (scaffold, RAG custom, system prompts) → devin irelevante
+- **Skill nou 2026:** Capacitatea de a anticipa cum un model mai inteligent schimbă workflow-ul ȘI de a te adapta
+
+---
+
+## 💬 Quote-uri Importante
+
+> "Is this instruction here because the model needs it or is it here because I needed the model to need it?"
+
+> "Ask for what you want in the end and explain why in plain language. And you don't need to elaborate on how to get there."
+
+> "Increasingly across 2026, this is the bitter lesson we have to learn. All of the way we have described process [...] doesn't matter anymore and what matters is the outcome."
+
+> "The art of prompting is evolving [...] the skill is evolving because the models are getting better. Increasingly the art of prompting is about **what you leave out**."
+
+> "We are moving toward a point where we want one eval gate at the end of the software process and it needs to check absolutely everything."
+
+> "Human talent will not [make up for it]. Like increasingly the whole point of human talent is to **simplify and get out of the way** so that AI can do its thing."
+
+> "Claude Mythos is coming. The inflection point is here. This is another one of those moments when you need to be able to catch the train before it leaves the station."
+
+> "When models get bigger, they force you to simplify."
+
+---
+
+## 💡 Idei & Aplicații pentru Marius
+
+### Immediate (Înainte de Mythos)
+1. **Audit prompt-urilor existente:**
+   - Ralph PRD generator - pot simplifica instrucțiunile?
+   - Cron jobs (rapoarte, coaching) - scot linii procedurale?
+   - System prompts pentru Echo - ce e hardcodat inutil?
+
+2. **Pregătire retrieval:**
+   - Organizare mai bună a memory/kb/
+   - Lăsa modelul să caute ce are nevoie vs a specifica eu rag logic
+
+3. **Simplificare workflow:**
+   - Unde specific prea mult HOW în loc de WHAT + WHY?
+   - Exemplu: procesare bonuri, email workflows
+
+### Post-Mythos (când se lansează)
+1. **Test security:** Ruleaza Mythos împotriva ROA repos pentru vulnerabilities
+2. **Upgrade Ralph:** Outcome specs mai clare, mai puțin procedural
+3. **Re-evaluate Max plan:** $200/lună worth it pentru acest nivel de capacitate?
+
+### Strategic
+1. **Skill de cultivat:** Anticipare cum modele mai bune schimbă workflow + adaptare rapidă
+2. **Mindset shift:** Outcome + evals, nu proces descris
+3. **Tool definitions:** Investește în tool descriptions clare pentru agenți
+
+---
+
+## 📚 Sursă
+
+Video produs de un creator AI/productivity focus. Ton: urgent dar practic, cu exemple concrete din customer support, software dev, knowledge work.
+
+**Credibilitate:** References concrete (Ghost repo vulnerability, stock market drops pentru cyber security firms, confirmări de la Anthropic).
+
+**Aplicabilitate:** Înaltă - checklist acționabil pentru pregătire.

memory/kb/youtube/2026-04-02-23-ai-trends.md

@@ -0,0 +1,207 @@
+# 23 AI Trends keeping me up at night
+
+**URL:** https://youtu.be/lyqk7zxbCKs  
+**Durată:** 31:36  
+**Salvat:** 2026-04-02  
+**Tags:** @work @growth
+
+## TL;DR
+
+Prezentare detaliată a 23 de tendințe AI care definesc momentul actual ca fiind extraordinar de favorabil pentru a construi startup-uri. Accentul cade pe viteza de execuție (companii create în ore, nu luni), trecerea de la modele per-seat la pricing bazat pe rezultate, verticalizarea AI (agenți care înlocuiesc headcount, nu doar software), și construirea de "ambient businesses" (afaceri care rulează autonom cu agenți). Mesaj central: fereastra de oportunitate este asimetrică dar limitată (12-24 luni) - cine începe acum câștigă avantaje de first-mover în nișe și audiențe încă underpriced.
+
+## Puncte Cheie
+
+### 🚀 Viteza de Execuție (The 1-Hour Company Stack)
+- **2015 timeline:** idee → hiring devs (luni) → MVP (luna 3) → launch → first revenue (luna 12)
+- **2026 timeline:** idee (9 AM) → build (9:15 AM) → product (9:45 AM) → first customer (10 AM) → iterate (lunch)
+- **Enablers:** Claude Code, Codex, Google AI Studio + audiență existentă + stripe integration
+- **IdeaBrowser.com** - idei validate, gata de vibe-coding
+
+### 🤖 Ambient/Autonomous Businesses
+- Afaceri care rulează cu **zero sau foarte puțin input uman zilnic**
+- Agenți care monitorizează piața, identifică oportunități, execută, handle customer service
+- **Check-in o dată la câteva zile** în loc de micromanagement
+- Predicție: vom vedea ambient businesses cu **seven-eight figures** în curând
+
+### 📈 Agent Economy Timeline
+- **2009-2015:** App Store era (humans operated apps)
+- **2015-2024:** API economy (devs wired APIs together)  
+- **2025-2030:** Agent economy (agents discovering & hiring other agents on the fly)
+- **Gartner stat:** 20% of commerce by 2030 will be **agent-to-agent / machine-to-machine**
+- **Market size:** $52 billion by 2030 (31,000 agent skills today, majoritatea slabe)
+
+### 💰 Pricing Revolution: Seat-Based → Outcome-Based
+- **Vechiul model:** $100/seat/month × 10 seats = $1,000/mo (valoare unclear)
+- **Noul model:** $1.50/resolved ticket = pay only for results  
+- **Zendesk** deja face asta; **83% of AI-native SaaS** switched to outcome pricing
+- **De ce:** Agenții fac munca efectivă → clienții plătesc pentru **labor replacement**, nu IT budget
+- **Gartner:** 40% of enterprise SaaS shifts to outcome-based by 2030
+- **Oportunitate:** "Someone's going to build a billion-dollar business doing nothing but converting legacy SaaS to outcome pricing"
+
+### 🏢 Vertical AI > Vertical SaaS
+- **Vertical SaaS:** Captures fraction of IT budget, sells software licenses, $10-100M outcomes
+- **Vertical AI:** Taps directly into **labor P&L**, agents do the work, **10x bigger TAM**
+- **YC prediction:** 300+ unicorns in vertical AI this decade
+- **Boring gold mines:** Insurance (30-year actuary tables), legal, logistics, elder care, government, accounting, construction
+- **Strategie:** Pick a **sub-niche wedge**, avoid big crowded categories
+
+### 🧠 Founder-Agent Fit (vs Founder-Market Fit)
+- **Noul skill:** Can you orchestrate a fleet of agents towards your goal?
+- **Analogy:** Film director - nu ții camera, nu acționezi, nu scrii partitura → **obții performanțe** din actori (agenți)
+- **Ghost Team Org Chart:** About page cu 2-3 oameni + sales agents, content agents, customer support agents (cu nume, personalități, eventual imagini)
+
+### 💡 100 True Fans (not 1,000)
+- **Kevin Kelly:** 1,000 true fans = viable business
+- **AI age:** **100 true fans** = real business (agenții reduc costurile dramatic)
+- **Micro-monopoly math:**
+  - 5,000 engaged niche audience
+  - Build custom app in 48h
+  - 100 customers × $50/month = $5,000/mo
+  - Run with agents → **$60k/year profit** for one person
+  - Replicate & incubate multiple
+
+### 📊 SaaS Graveyard vs What Survives
+**Will die:**
+- Generic CRMs (agenții fac mai bine)
+- Basic analytics dashboards (AI generates insights on demand)  
+- Template marketplaces (AI generates custom instantly)
+- Scheduling tools (agents handle calendars natively)
+- Basic customer support chatbots (already being replaced)
+
+**Will survive:**
+- Vertical workflow tools that pivot to agents
+- Infrastructure & data moats
+- Companies moving toward outcome-based pricing
+
+### 🌐 Scarcity Flip: Execution → Judgment
+**Commoditized by AI:**
+- Generic content, basic design, data entry, routine analysis
+
+**Scarce & Premium:**
+- **Creative judgment** (human-led decision making)
+- **Human-made crafts** (Porsche "100% human-made" campaign = premium positioning)
+- **Physical experiences** (IRL economy)
+- **Original weird thinking** (LLMs can't be weird)
+- **Proprietary data**
+
+**Premium Stack:**
+1. **Most premium:** Human-made, no AI involved (organic-style certification)
+2. **Premium:** AI-assisted but human-led (human taste + AI speed)
+3. **Commodity:** Fully AI service
+4. **Race to zero:** Pure automation
+
+### 🏡 IRL/Experience Economy
+- **When digital is infinite**, scarcity shifts to **physical presence with other humans**
+- **Opportunities:** Karaoke bars, escape rooms, immersive theater, co-working, live music
+- **Already accelerating** - huge opportunity in physical experiences
+
+### ⚠️ Agent Attack Surface (the scary part)
+- **Threats:** Prompt injections, poison context windows, malicious MCP services, agent-to-agent manipulation, permission escalation, compromised training data
+- **Agent Injection > Phishing:**
+  - Phishing (2010): trick human → click bad link (billions lost/year)
+  - Agent Injection (2026): trick AI via hidden instructions → targets context windows, agent autonomy is the vulnerability  
+  - **Potentially far bigger** (agents have system access + make autonomous decisions)
+- **Palo Alto Networks** documented real-world agent injection attacks
+- **Opportunity:** Cybersecurity software for agent defense
+
+### 🔐 Agent Permission Stack
+**What can your agent:**
+- **Access:** files, emails, calendars, bank accounts (people giving $5k for trading!)
+- **Remember:** conversations, personal data, business data  
+- **Do:** send emails, make purchases, modify code, delete data
+- **Share:** with other agents or third parties
+
+**Digital hygiene:** Quarterly agent cleanses - review permissions like app access
+
+### ⏰ The Asymmetric Window (12-24 Months)
+**What you need:**
+- API key, some prompts, a tweet, niche audience (100-5,000)
+
+**What you can get:**
+- Business running 24/7
+- 95% margins (realistic: 60-80%)  
+- Compounding distribution
+- Zero or few employees
+
+**Timeline:**
+- **Now:** Build cost = zero, niches wide open, audiences underpriced
+- **12 months:** Competition catches up, best niches claimed, tools crowded
+- **24 months:** Window narrows - early builders own moats (data, network, brand, trust)
+
+**Message:** "People keep waiting for things to settle down. Things are NOT settling down. This is the new normal."
+
+### 🌍 Build in Public (still valuable)
+- **Old advice:** "Build in public!"  
+- **Current pushback:** "Don't build in public - you're inviting competition"
+- **Reality:** **Benefits outweigh cons** when:
+  - Your followers/audience ARE your customers
+  - Community votes on what you're building
+  - Ship updates in 1-5 days (users become co-builders)
+  - Increases trust + distribution compounds
+- **Moat:** In a world where you can fork businesses like GitHub repos, **community engagement** is a huge competitive advantage
+
+### 🔄 Forking Businesses (like GitHub repos)
+- In AI age, **copying other people's businesses is fast & easy**
+- **Defensive strategy:** Bring community in, make them feel part of building
+- **Community = moat** when execution speed is commoditized
+
+## Quote-uri Memorabile
+
+> "The fact that you can create a company in a day... I don't want to just build one company, try it for six months. I want to just create a culture of and a machine that I'm creating multiple companies trying different things."
+
+> "Unlike SaaS which captures IT budget, vertical AI replaces headcount and that's just a 10x bigger total addressable market."
+
+> "Someone's going to build a billion-dollar business doing nothing but converting legacy SaaS to outcome pricing."
+
+> "I think it's more like the 100 true true fans because agents are cutting your cost so dramatically that 100 people paying you is a real business."
+
+> "The value is going to migrate from execution to judgment."
+
+> "When digital is infinite and AI generated, scarcity shifts to physical presence with other humans."
+
+> "People keep waiting for things to settle down. Things are NOT settling down. This is the new normal."
+
+> "This is the most asymmetric time to be building a startup."
+
+> "Users are basically becoming co-builders and that just increases trust and distribution compound."
+
+> "You're just learning as you go. You're building as you go with momentum every day better than the next. What an incredible time to be building."
+
+## Oportunități de Business Menționate
+
+1. **Glassdoor of AI Agents** - marketplace/social network for agent reputation
+2. **Agent cybersecurity** - defend against prompt injection, context poisoning
+3. **Constellation Software for Vertical AI** - hold 500+ AI-native agent businesses in boring verticals
+4. **Legacy SaaS → Outcome Pricing Conversion** - billion-dollar opportunity
+5. **IRL/Experience Economy** - karaoke, escape rooms, immersive theater, co-working
+6. **Micro-monopolies** - 100-customer niche apps running on agents
+7. **Agent Skills Marketplace** - quality skills (current 31k mostly garbage)
+
+## Idei de Acțiune pentru Marius
+
+### Imediat Acționabil (Low Effort, High Impact)
+1. **🎯 Review pricing ROA/roa2web:** Există componente care pot trece pe outcome-based? (ex: rapoarte generate, notificări trimise, validări ANAF automatizate)
+2. **📝 Identifică 1-2 pain points repetitive** pe care le ai cu clienții → candidați pentru micro-tool agent-driven (100 true fans model)
+3. **🤖 Testează "1-hour company stack"** cu o idee mică din IdeaBrowser.com - exercițiu de learning, nu neapărat production
+4. **🔐 Agent permission audit:** Revizuiește ce acces au toolurile tale AI (Claude Code, etc.) - digital hygiene
+
+### Mediu Termen (Strategice, Planning)
+5. **💡 Vertical AI wedge for ROA ecosystem:** Ce sub-nișă din accounting/legal/ERP poți domina cu agenți? (ex: "ANAF compliance automation for small accounting firms")
+6. **👥 Ghost team planning:** Ce parts din workflow-ul ROA pot fi automatizate cu agenți documentați public? (marketing, customer support, onboarding)
+7. **📊 SaaS graveyard check:** Care din toolurile pe care le plătești lunar ar putea fi înlocuite cu soluții AI-native outcome-based?
+
+### Long Term (Transformațional)
+8. **🏢 Ambient business design:** Cum ar arăta ROA/roa2web ca "ambient business" unde check-in-ul e o dată la 2-3 zile, nu zilnic?
+9. **🌍 IRL component:** Experiențe fizice pentru clienți ROA? (workshops, events, training - scarcity shift)
+10. **🔄 Community co-building:** Poate un Discord/grup privat pentru clienți ROA unde votează features noi pentru roa2web? (build in public, distribution moat)
+
+## Link-uri Menționate
+- **IdeaBrowser.com** - validated startup ideas
+- **Paperclip** - open-source agent orchestration (tutorial disponibil)
+
+---
+
+**Meta-Insight pentru Echo:**  
+Acest video reinforces ideea că Marius ar trebui să gândească mai puțin "cum să implementez totul manual" și mai mult "ce pot automatiza cu agenți și cum pot scala prin replicare de micro-businesses". 80/20 mindset perfect aplicabil aici: identifică cele mai boring, repetitive workflows din ROA/roa2web și transformă-le în "ambient components" care rulează autonom.
+
+Pricing-ul outcome-based e direct relevant pentru roa2web - în loc de "€X/lună pentru interfață web", gândește "€Y per raport generat" sau "€Z per notificare ANAF procesată". Asta face sales-ul mult mai ușor (no brainer pentru clienți când plătesc doar pentru value delivery).