1. AGI and the Current State of AI
- AGI should be viewed as a transition period, not a single moment — and it has already begun as of early 2025.
- Frontier models already exceed human capability in some domains (math, coding, biology).
- Hallucination occurs when models lack context or proprietary data.
"AGI is a work in progress."
2. Verifiable Reward Functions and Reinforcement Learning
- Domains with verifiable reward functions (math, coding, biology) see rapid AI advancement.
- The key to AI progress: turning non-verifiable into verifiable through closed-loop feedback systems.
- AlphaEvolve uses hallucination as an innovation tool — exploring possibility spaces through diverse attempts.
3. Evolution of AI Coding Tools
- Cursor evolved from code assist to agentic workflow, building self-models from user data and creating a virtuous cycle of execution speed and market share.
- Senior engineers benefit most — AI amplifies their full-stack capabilities.
- Junior engineers face reduced hiring, but gain new entrepreneurial opportunities.
"Direction-setting, strategy, and evaluation are now far more important skills than implementation."
4. AI-Native Company: Three Essential Elements
- Leader: Connects AI and domain
- AI Engineer: Rapidly applies latest AI trends
- Internal Customer: Domain experts (non-engineers)
Alignment between these three takes approximately 2 years.
5. Stages of Company Evolution
AI-assisted → AI-driven → Autonomous → Self-evolving company.
Human roles shift toward evaluator and strategist. Prompting and evaluation become the primary human functions.
"At the ideal end state, core processes run on AI, and humans handle direction, strategy, prompting, and evaluation."
6. Ethics, Human Purpose, and Reality
- AI automation may lean toward replacement over augmentation for efficiency.
- New roles: AX (AI Transformation) talent; prompting and evaluation competencies.
"If you can't clearly envision yourself in a role that involves good prompting or good evaluation, you're in danger."
7. Replication and Domain Innovation
- The "Cursor for X" model — replicate the structure for each domain.
- Alignment of the three essential elements (leader, AI engineer, internal customer) is critical but takes 2-3 years.
"This change has a 99.8% probability of happening, so starting early is the answer."
Key Terms: AGI Transition, Verifiable Reward Function, Closed-Loop Feedback, Agentic Workflow, Cursor for X, Proprietary Data, Vertical AI, AI-Native Company, Prompting & Evaluation, Autonomous Company