Reid Hoffman Leaves Microsoft's Board to Go "Founder Mode" with Startup Manus

This article is part of the daily AI Coding newsletter. For more in-depth technical analysis and code examples, subscribe to our newsletter.

Introduction

Reid Hoffman, the legendary entrepreneur often dubbed “the rules-setter of Silicon Valley,” has made a striking career move—leaving Microsoft’s board after a decade of service to join AI drug discovery startup Manus as a co-founder. This decision marks not only a new chapter in Hoffman’s personal career but also profoundly reflects the rise of “Founder Mode” as an emerging management paradigm in today’s AI industry.

---

1. Background and Relationship Mapping

1.1 Reid Hoffman’s Position in Silicon Valley

Reid Hoffman is one of the most influential figures in Silicon Valley, with a legendary track record:

Company/Role	Period	Significance
PayPal Early Member	2000-2002	Co-founded with Peter Thiel, Elon Musk
LinkedIn Co-founder & CEO	2002-2008	IPO in 2008, acquired by Microsoft for $26.2B in 2016
Greylock Partners Partner	Investment career	Invested in Airbnb, Facebook, and other tech giants
OpenAI Early Investor	2015-2023	Witnessed the rise of GPT series models
Microsoft Board Member	2016-2026	Ten-year tenure, witnessed Microsoft’s AI transformation

1.2 Key Transaction Timeline

2016 ── Microsoft acquires LinkedIn for $26.2 billion
          │
          └──► Hoffman joins Microsoft's board
                    │
2019 ───────────────┴─── Microsoft invests first $1 billion in OpenAI
                              │
2023 ──────────────────────────┴─── Hoffman exits OpenAI board due to conflicts
                                        │
2023 ───────────────────────────────────┴─── Microsoft signs $650M "non-acquisition acquisition" with Inflection AI
                                                  │ (hires co-founder Mustafa Suleyman)
                                                  │
2026 ────────────────────────────────────────────┴─── Hoffman leaves Microsoft board, joins Manus

1.3 The Cumulative Effect of Conflict of Interest

Hoffman’s cross-board memberships across multiple AI-related entities created governance challenges:

1. LinkedIn Acquisition: As the founder of the company Microsoft acquired, Hoffman received substantial proceeds
2. OpenAI Investment: Hoffman simultaneously served on both OpenAI’s and Microsoft’s boards, creating concerns about potential conflict of interest as the two companies deepened their partnership
3. Inflection AI Deal: Microsoft paid $650 million to “employ” the Inflection team, while Hoffman was an investor in Inflection

This layered relationship structure ultimately led Hoffman to step down from the OpenAI board in 2023.

---

2. Manus: A Emerging Force in AI Drug Discovery

2.1 Company Overview

Item	Details
Company Name	Manus
Founders	Reid Hoffman (Co-founder & Chairman), Dr. Siddhartha Mukherjee (CEO)
Domain	AI-powered drug discovery
Funding	Over $50 million (multiple seed rounds)
Key Investors	Reid Hoffman, General Catalyst
Technical Core	Move 37 AI—AI that surpasses human creativity in chemistry

2.2 Move 37 AI Technical Deep-Dive

Manus’s core technology is called Move 37 AI, a name that alludes to AI’s “37th move” in the game of Go—an analogy to AlphaGo’s famous move 37 that revolutionized go theory. In Go, AI discovered a move that human players had never imagined, fundamentally changing how the game is played.

Core Capabilities of Move 37 AI:

# Pseudocode example: Move 37 AI's molecular generation logic
class Move37MolecularGenerator:
    """
    Move 37 AI Core Algorithm
    Goal: Surpass human creativity in molecular design
    """
    
    def __init__(self, target_cancer_type: str):
        self.target = target_cancer_type
        # Pre-trained on millions of known molecular structures
        self.molecule_encoder = MolecularTransformer()
        # Curiosity-driven exploration via reinforcement learning
        self.curiosity_drive = CuriosityDrivenOptimizer()
    
    def generate_novel_compound(self) -> MolecularStructure:
        # Stage 1: Target-conditional generation
        candidate = self.molecule_encoder.generate(
            target=self.target,
            constraints=ChemicalConstraints()
        )
        
        # Stage 2: Reinforcement learning optimization (beyond human intuition)
        optimized = self.curiosity_drive.optimize(
            molecule=candidate,
            reward_function=self.efficacy_score
        )
        
        # Stage 3: Synthesis feasibility validation
        if self.synthetic_accessibility(optimized) > 0.7:
            return optimized
        return self.generate_novel_compound()

2.3 Traditional vs AI-Powered Drug Discovery

Dimension	Traditional Drug Discovery	Manus Move 37 AI
Development Timeline	10-15 years	Expected 50%+ reduction
Cost	$2-3 billion per drug	Expected 80%+ reduction
Molecular Library Scale	Millions of compounds	Billions of virtual compounds
Target Discovery	Manual screening	AI-driven prediction
Creativity	Limited by human expert experience	Surpasses human intuition

---

3. The Rise of “Founder Mode”

3.1 What is “Founder Mode”?

“Founder Mode” is a management philosophy contrasting with professional management specialization, emphasizing the founder’s absolute control over company direction. This concept was reintroduced by Y Combinator founder Paul Graham in 2023 and sparked widespread discussion in Silicon Valley’s startup community.

Core Characteristics of Founder Mode:

# Founder Mode vs Professional Management Mode
class CompanyManagementModel:
    
    def founder_mode(self):
        return {
            "decision_making": "Fast, founder-led",
            "product_vision": "Founder has clear control",
            "execution": "Cross-functional direct intervention",
            "hiring": "Only hire the best",
            "meetings": "Founder high-frequency participation",
            "innovation": "Tolerate risk, encourage breakthroughs"
        }
    
    def professional_mode(self):
        return {
            "decision_making": "Hierarchical approval, committee decisions",
            "product_vision": "Interpreted by professional managers",
            "execution": "Clear departmental divisions",
            "hiring": "Hire per budget",
            "meetings": "Delegated to teams",
            "innovation": "Avoid risk, optimize existing business"
        }

3.2 Why Does Hoffman Need “Founder Mode”?

Hoffman stated in a recent episode of his Possible podcast:

“One thing I realized in the past month is that we see Manus making such huge progress. I need to get back into founder mode.”

The Logic Behind This:

1. AI Startup Competition Pace: The AI field is advancing rapidly, requiring founder-level高频介入
2. Technical Uncertainty: Move 37 AI is still in early stages, requiring founder to control technical direction
3. Strategic Resource Allocation: Resource coordination with General Catalyst requires founder perspective
4. Investor Confidence: Hoffman’s full-time involvement is crucial for subsequent fundraising

---

4. Implications for the AI Coding Domain

4.1 AI Agent Applications in Drug Discovery

The Manus case demonstrates successful AI Agent application in vertical domains. Here’s a simplified example of AI Agent design patterns in drug discovery scenarios:

"""
AI Agent Design Pattern in Drug Discovery
Reference: Manus's Move 37 AI Architecture
"""
from dataclasses import dataclass
from typing import List, Optional
from enum import Enum

class AgentState(Enum):
    ANALYZING = "analyzing"
    GENERATING = "generating"
    VALIDATING = "validating"
    SYNTHESIZING = "synthesizing"

@dataclass
class MolecularStructure:
    smiles: str  # Simplified Molecular Input Line Entry System
    binding_affinity: float
    toxicity_score: float
    synthetic_score: float

class DrugDiscoveryAgent:
    """
    Drug Discovery AI Agent
    Simulating Manus Move 37 AI's Core Process
    """
    
    def __init__(self, target_disease: str):
        self.target_disease = target_disease
        self.state = AgentState.ANALYZING
        self.knowledge_base = self._load_biomolecular_knowledge()
    
    def _load_biomolecular_knowledge(self):
        """Load biomolecular knowledge base"""
        return {}
    
    def discover_drug(self, target_protein: str) -> List[MolecularStructure]:
        """
        Core Discovery Process:
        1. Analyze target structure
        2. Generate candidate molecules
        3. Validate pharmacokinetic properties
        4. Assess synthesis feasibility
        """
        candidates = []
        
        # Stage 1: Target analysis
        protein_structure = self._analyze_protein(target_protein)
        
        # Stage 2: Generate candidate molecules (surpassing human creativity)
        raw_candidates = self._generate_molecules(protein_structure)
        
        # Stage 3: Multi-round optimization
        for candidate in raw_candidates:
            if self._passes_filters(candidate):
                candidates.append(candidate)
        
        return candidates[:10]
    
    def _analyze_protein(self, protein: str) -> dict:
        """Use AlphaFold-like models to predict protein structure"""
        return {"structure": "predicted", "binding_sites": []}
    
    def _generate_molecules(self, protein_info: dict) -> List[MolecularStructure]:
        """Use generative AI to generate new molecules"""
        return []
    
    def _passes_filters(self, molecule: MolecularStructure) -> bool:
        """Multi-dimensional filtering"""
        return (
            molecule.binding_affinity > 0.8 and
            molecule.toxicity_score < 0.2 and
            molecule.synthetic_score > 0.7
        )

4.2 Analogies in the AI Coding Domain

The Manus and Hoffman collaboration model highly resonates with development trends in the AI Coding domain:

Domain	Drug Discovery (Manus)	AI Coding (e.g., Cursor, Copilot)
Core AI Capability	Molecular generation and optimization	Code generation and optimization
Human Creativity Role	Provide biological insights	Provide architectural design decisions
AI’s Role	Explore chemical space	Explore code space
Validation Stage	Wet lab validation	Unit test validation
Breakthrough Potential	New drug categories	New programming paradigms

---

5. Microsoft’s Deep Connection with the AI Industry

5.1 Microsoft’s AI Investment Portfolio

Hoffman’s departure from Microsoft’s board actually reflects the breadth and complexity of Microsoft’s AI布局:

Microsoft AI Investment Portfolio (2015-2026)
├── OpenAI (2019 $1 billion investment, cumulative >$13 billion)
│   ├── Exclusive cloud provider for GPT series models
│   └── Core of Azure OpenAI Service
├── Inflection AI (2023 $650M "non-acquisition acquisition")
│   └── Hired co-founder Mustafa Suleyman and team
├── LinkedIn (2016 $26.2 billion acquisition)
│   └── Core asset for Microsoft's-to-B strategy
└── GitHub (2018 $7.5 billion acquisition)
    └── Copilot's carrier platform

5.2 Hoffman’s “Exit” and “Entry”

Departure	Join	Strategic Logic
Microsoft Board	Manus Co-founder	From investor/advisor to一线创业者
OpenAI Board	-	Eliminate conflict of interest, focus on new project
Inflection AI Investor	-	Shift to more breakthrough-oriented Manus
LinkedIn Founder Identity	-	History completed, seeking new chapter

---

6. Technical Deep-Dive: Why is AI Drug Discovery Exploding Now?

6.1 Three Technical Pillars

AI drug discovery breakthroughs are built on three major technological advances:

1. Pretrained Large Models Applied to Molecular Representation

# Evolution of molecular representation learning
"""
Traditional method: SMILES string representation
        ↓
Evolutionary method: Molecular Graph Neural Networks (GNN)
        ↓
State-of-the-art: Molecular Language Models (Molecular LM)
         - Similar to LLM predicting tokens, predicting molecular properties
         - Captures substructure-level semantic information
"""

class MolecularLanguageModel:
    """
    Molecular Language Model Architecture
    Input: SMILES or molecular graphs
    Output: Molecular property prediction, new molecule generation
    """
    
    def __init__(self, model_size="large"):
        self.encoder = self._build_molecular_encoder()
        self.decoder = self._build_molecular_decoder()
        # Pre-trained on PubChem, ChEMBL and other large-scale chemical databases
        self.pretraining_data = "billions of molecules"
    
    def generate(self, target_property: dict) -> str:
        """Generate new molecules based on target properties"""
        pass

2. Reinforcement Learning-Driven Molecular Optimization

Traditional optimization methods are inefficient in chemical space, while reinforcement learning can:

• Efficiently explore billions of virtual compound libraries
• Simultaneously optimize multiple objectives (efficacy, toxicity, drug-likeness)
• Discover molecular modification paths that human experts struggle to conceive

3. Breakthroughs in Protein Structure Prediction Models like AlphaFold

AlphaFold 2 (2020) and subsequent versions completely changed the target analysis process:

Traditional Process	AlphaFold Era
Protein crystal structure determination (years)	Predicted structure (hours)
Requires expensive experimental equipment	Pure computation
Only covers known structures	Theoretically covers all proteins

6.2 Moore’s Law Applied to Drug Discovery

The cost curve of AI drug discovery is following in the footsteps of Moore’s Law:

Cost ($)
^
|  Traditional Pharma
|  │
|              │ AI Pharma (exponential decline)
|              │ │
|_______________│_______________│________________► Time
  2010          2020          2030 (expected)
 
Cost per drug: $2-3 billion → expected <$500 million

---

7. Industry Impact and Future Outlook

7.1 Implications for the AI Startup Ecosystem

1. Vertical specialization is key to AI落地: Manus focuses on cancer treatment, avoiding direct competition with major pharmaceutical companies
2. Value of Founder Mode in the AI era: High technical uncertainty requires rapid founder decision-making
3. “Non-Acquisition Acquisition” model: Microsoft’s deal with Inflection may be replicated by more major companies

7.2 Potential Challenges and Risks

Risk Category	Specific Risk	Mitigation Strategy
Technical Risk	Clinical trial failure	Partner with traditional pharma
Regulatory Risk	FDA approval	Early regulatory affairs布局
Business Risk	Competition from Big Pharma	Focus on niche segments
Talent Risk	Shortage of AI + biology talent	University partnerships

---

Conclusion

Reid Hoffman’s departure from Microsoft’s board to join Manus marks a new phase in AI entrepreneurship. This event is not merely a personnel change but a renewed confirmation of “Founder Mode’s” value in the AI era.

Key Takeaways:

1. Personal Level: Hoffman’s career trajectory reflects how top investors judge the direction of AI transformation
2. Technical Level: Move 37 AI represents the latest progress in AI drug discovery—surpassing human creativity in chemical exploration
3. Business Level: Manus adopts “Co-founder + Professional CEO” model, balancing entrepreneurial passion with operational expertise
4. Industry Level: AI vertical落地 (drug discovery, code generation, etc.) is accelerating

For practitioners in the AI Coding domain, the Manus case provides an important reference: when AI demonstrates creativity surpassing humans in specific domains (such as chemistry, code), “Founder Mode” management philosophy and rapid technical roadmap iteration will become key to competitiveness.

---

Related Topics: AI Agent, MCP (Model Context Protocol), AI Drug Discovery, Founder Mode, Generative AI