Deal Velocity, Not Billable Hours: How Crosby Uses AI to Redefine Legal Contracting

AI-Powered Legal Contract Automation

Crosby is an AI-first law firm that focuses entirely on contracts with the goal of automating human negotiations. The company operates on the theory that they can help people agree on terms faster through automated contract processes.

Core Business Model:

1. Contract-Focused Approach - Specializes exclusively in contract work across all types (leases, offer letters, business transactions)
2. Automation Strategy - Replaces traditional lawyer workflows with AI agents to close contracts faster
3. In-House Integration - Built as an actual law firm to understand how lawyers work before automating their processes

Key Differentiators:

• Speed Focus: Achieves contract turnaround times under an hour with strict SLA commitments
• Operational Metrics: Tracks human touchpoints, turnaround time, and automation success rates
• Quality Assurance: Maintains lawyer oversight while leveraging AI for efficiency
• Customer Trust: Serves fast-growing companies like Cursor that need both speed and excellence

The approach combines the most promising aspects of AI automation with the credibility and expertise that clients expect from traditional legal services.

Timestamp: [1:28-7:59]

Strategic Structure for Innovation

The decision to build a law firm rather than legal software stems from structural advantages that enable better innovation and market positioning.

Structural Advantages:

1. Investment Flexibility - Traditional law firm partnerships can't invest in speculative technology because they can't sell equity and only partners can take recourse loans
2. Human Capital Premium - Legal services market puts huge premium on human expertise, similar to VC firms investing in people
3. Innovation Barriers - Law firms historically haven't been able to innovate in task automation due to partnership constraints

Market Evolution:

• 30-Year Shift: Growing recognition that specialized legal work can be offloaded to technology
• Recent Acceleration: Last three years have shown dramatic changes in what's possible with AI
• Structural Solution: Getting the business structure right enables rapid iteration cycles and real innovation

Precedent and Learning:

• Atrium's Proof: Company proved this model 5 years ago, demonstrating viability
• Limited Attempts: Only 2-3 companies (Atrium, ClearSpire) have meaningfully tried this approach
• Unique Timing: Current AI capabilities make huge chunks of qualitative legal work delegatable to machines

Timestamp: [2:05-3:31]

Side-by-Side Integration Model

Crosby creates unique feedback loops by having domain experts (lawyers) work directly alongside engineers in an integrated environment.

Physical Integration:

• Desk Arrangement: Office has staggered seating - lawyer, engineer, lawyer, engineer
• Collaboration Incentives: Physical proximity designed to encourage real-time collaboration
• Feedback Cycles: Creates immediate opportunities for iteration and improvement

Operational Benefits:

1. Beyond Benchmarks - More than just evaluation metrics; actual product usage and user research
2. Workflow Understanding - Engineers experience critical workflows and identify biggest gaps firsthand
3. Quality Control - Real-time feedback on contract reviews, redlines, and word choice accuracy

Product Development Process:

• Instrumentation Focus: Operational heavy business requires detailed metrics tracking
• Touch Point Analysis: Monitor every step from contract intake to completion
• Automation Prioritization: Identify highest-value processes to automate based on real usage data

Customer Impact Measurement:

• Leading Indicators: Product velocity, human touchpoints, turnaround time
• Lagging Indicators: SLA compliance, on-time delivery, customer satisfaction
• Quality Metrics: Ensuring automated processes meet client standards for excellence and speed

Timestamp: [3:36-7:06]

Credence Goods and Professional Credibility

The law firm structure addresses a fundamental challenge in legal services - clients need expert validation to assess quality.

Credence Good Challenge:

• Definition: Legal services are "credence goods" - you only know quality after consumption
• Expert Validation Required: Even sophisticated CEOs often can't assess legal work quality by themselves
• Trust Factor: Clients want assurance that a qualified lawyer is reviewing their work

Market Accessibility:

1. Client Comfort: Companies are excited about AI-enabled services but need familiar structure
2. Professional Standards: Law firm designation provides immediate credibility and trust
3. Quality Assurance: Combines AI efficiency with lawyer oversight for optimal results

Competitive Advantage:

• Best of Both Worlds: Most promising AI capabilities with traditional professional credibility
• Market Positioning: Accessible to companies wanting to try AI services within familiar framework
• Risk Mitigation: Clients get innovation benefits without sacrificing professional standards

Timestamp: [7:19-7:59]

Essential Insights:

1. AI-First Law Firm Model - Crosby operates as an actual law firm focused on contract automation, achieving sub-hour turnaround times
2. Structural Innovation - Building as a law firm rather than software company enables investment in technology while maintaining client trust
3. Integrated Collaboration - Lawyers and AI engineers work side-by-side with staggered desk arrangements to create immediate feedback loops

Actionable Insights:

• Traditional law firm partnerships can't innovate due to equity and loan restrictions, creating opportunity for new structures
• Physical proximity between domain experts and engineers accelerates product development beyond traditional benchmarks
• Legal services as "credence goods" require professional credibility even when leveraging cutting-edge AI automation
• Operational metrics like human touchpoints and SLA compliance are critical for scaling AI-enabled professional services

Timestamp: [0:46-7:59]

Companies & Products:

• Atrium - Proved the law firm + technology model 5 years ago, demonstrating structural viability
• ClearSpire - Early attempt at combining legal services with technology innovation
• Cursor - Fast-growing client company that uses Crosby's contract services

Concepts & Frameworks:

• Credence Goods - Economic concept where quality can only be assessed after consumption and requires expert evaluation
• Partnership Structure vs C-Corp - Business structure comparison affecting ability to invest in long-term technology
• SLA (Service Level Agreement) - Performance metrics used to measure contract turnaround time and service quality

Timestamp: [0:46-7:59]

Pricing Innovation in Legal Services

Crosby made an early decision to completely eliminate billable hours, which was both dramatic for the legal industry and an obvious way to align incentives within the company. This approach serves as an interesting wedge to get people's attention while fundamentally changing how legal work is priced.

Current Pricing Model:

• Per-document billing instead of hourly rates
• Predictive pricing from time zero for each piece of work
• Value-aligned pricing that matches the client benefit received

Why This Innovation Matters:

1. Historical Context - The billable hour has only been popular since the 1950s, and people have been predicting its death for 70 years
2. Efficiency Paradox - Making hourly work more efficient through AI doesn't help clients if they're still paying by the hour
3. Predictive Capability - The ability to predict exactly how long contract work will take enables fixed pricing

Pricing Complexity Factors:

• Negotiation rounds: Predicting how many times a contract will go back and forth (5 times, 3 times, 2 times)
• Document complexity: From simple 2-page NDAs to complex 15-page MSAs that are 80x more complex
• Risk assessment: Balancing client needs with proper legal protection

Timestamp: [8:12-9:49]

Contract Categories and Complexity Levels

Crosby currently focuses on three main types of contracts that every B2B tech company deals with regularly, arranged by increasing complexity levels.

Primary Contract Types:

1. NDAs (Non-Disclosure Agreements)

• Complexity: Lowest level, typically 2 pages
• Use case: Standard confidentiality protection
• Volume: High frequency for growing companies

2. MSAs (Master Service Agreements)

• Complexity: Significantly higher, typically 15 pages
• Relative complexity: Approximately 80x more complex than NDAs
• Terms: Many more negotiation points and variables

3. DPAs (Data Processing Agreements)

• Complexity: Similar to MSAs in sophistication
• Context: Essential for B2B SaaS companies handling data

Client Examples and Volume:

• High-growth companies like Cursor, Clay, and Unify are signing dozens of contracts per day
• Risk balance: Need to close deals quickly while protecting the business from excessive risk
• Scale challenge: Manual lawyer review becomes a bottleneck at this volume

Complexity Comparison:

• NDAs: 2 pages, basic terms
• MSAs: 15 pages, 80x more complex
• Merger agreements: 1000x more complex (future expansion area)

Timestamp: [10:03-11:12]

The Mental Models and Heuristics of Legal Decision-Making

Lawyers today rely heavily on internalized mental models and ethereal knowledge that exists primarily in their minds, creating inconsistencies and subjective decision-making in contract negotiations.

Current Lawyer Decision Process:

1. Risk Prediction - Analyzing all contract terms to determine what should and shouldn't be agreed to
2. Mental Model Calibration - Using internalized frameworks of what's "safe" and "unsafe" for specific businesses
3. Market Benchmarking - Relying on general consensus of what's considered "market standard"

Key Challenges with Human-Only Approach:

• Subjectivity: Two reasonable lawyers will disagree on what terms are acceptable
• Ethereal Knowledge: Market standards exist only in lawyers' minds, not in quantifiable data
• Inconsistency: Guidelines and benchmarks are general rather than data-driven

The Public Good Problem:

• Private Actors, Public Infrastructure: Lawyers are essentially private actors building public legal infrastructure
• Collective Knowledge Gap: All negotiated contracts contribute to a public domain of "reasonable risk allocation"
• Lack of Quantification: No statistical or metric-based approach to determine what's truly reasonable

Crosby's Quantitative Innovation:

• Data-Driven Benchmarks: Converting lawyer intuition into actual probabilistic numbers
• Statistical Risk Assessment: Providing metrics rather than "best guesses" for contract terms
• Market Data Analysis: Using real negotiation data to inform decisions

Timestamp: [10:54-12:07]

From Vibes to Data-Driven Legal Decisions

Crosby transforms the subjective "vibes and heuristics" inside lawyers' heads into concrete, probabilistic numbers that can guide contract negotiation decisions with statistical backing.

Quantification Process:

1. Probabilistic Analysis - Converting gut feelings into actual probability percentages
2. Market Data Integration - Using real negotiation outcomes to inform recommendations
3. Risk Lever Identification - Understanding specific mechanisms for allocating business risk

Concrete Example - Governing Law:

• Traditional approach: Lawyer's intuition about which state law to accept
• Crosby's approach: Statistical probability of success for Delaware vs. California vs. New York governing law in specific negotiation contexts
• Data advantage: Only 50 states to analyze, making the dataset manageable

Key Innovation Areas:

• Human + Quantitative: Maintaining the inherently human aspect of agreement while adding statistical backing
• Market Intelligence: Dissecting actual market data to understand risk allocation levers
• Founder Education: Helping business leaders understand which legal levers they can actually control

Business Strategy Implications:

• PMF vs. Services Trap: Distinguishing between customer willingness to pay and actual product-market fit
• Automation Focus: Understanding what is truly automatable and repeatable
• Signal Recognition: Identifying specific signals and levers that can drive automation forward

Timestamp: [12:07-13:15]

The Architecture of Human-AI Collaboration in Legal Work

Crosby's approach focuses on context engineering for human lawyers first, then gradually expanding AI capabilities based on proven human workflows and decision-making patterns.

Current Workflow Architecture:

1. Context Engineering for Lawyers - Providing human lawyers with enhanced tools and building blocks
2. Manual Workflow Automation - Automating the repetitive tasks lawyers previously did manually
3. Behavioral Replication - Training language models to replicate specific human lawyer behaviors

The Context Revolution:

• Historical progression: From simple image classification (just the picture) to complex medical diagnosis (MRI + patient history + metadata)
• Language model advantage: Can process unlimited context to understand problems and make decisions
• Legal application: Providing comprehensive context about contracts, precedents, and business requirements

AI Model Training Philosophy:

• Individual alignment over general-purpose training
• Specific person replication rather than broad consensus
• Internal consistency focus - aligning with one lawyer's decision-making patterns
• Per-customer customization - Fine-tuning models for individual clients and their specific needs

Why Individual Alignment Works:

• Lawyer disagreement reality: Even two lawyers from the same firm will disagree on contract terms
• Consistency value: One person's internally consistent approach is more valuable than averaged consensus
• Accuracy improvement: Models perform better when aligned to specific individuals rather than general populations

Timestamp: [13:15-15:55]

Essential Insights:

1. Pricing Revolution - Crosby eliminated billable hours in favor of per-document pricing, aligning incentives and enabling predictive cost modeling for legal work
2. Quantified Legal Intelligence - Converting subjective lawyer intuition into statistical, data-driven recommendations for contract negotiations
3. Individual AI Alignment - Training language models to replicate specific lawyers' decision-making patterns rather than seeking broad consensus

Actionable Insights:

• Contract complexity scaling: NDAs (2 pages) to MSAs (15 pages, 80x more complex) to merger agreements (1000x more complex)
• High-growth company reality: Companies like Cursor, Clay, and Unify are signing dozens of contracts daily, creating automation opportunities
• Context engineering approach: Start by enhancing human lawyer capabilities, then gradually automate proven workflows
• Per-customer fine-tuning: Customize AI models for individual clients rather than using general-purpose solutions

Timestamp: [8:06-15:55]

People Mentioned:

• AlexNet researchers - Referenced in context of early image classification breakthroughs that parallel current language model development

Companies & Products:

• Cursor - High-growth company example signing dozens of contracts daily
• Clay - Fast-growing company dealing with high contract volumes
• Unify - Another example of rapidly scaling company with significant contract negotiation needs

Technologies & Tools:

• AlexNet - Early deep learning model used as comparison point for current language model capabilities
• Language Models/LLMs - Core technology for automating legal decision-making processes
• RLHF (Reinforcement Learning from Human Feedback) - General AI alignment technique mentioned in contrast to individual-specific alignment

Concepts & Frameworks:

• Billable Hour Model - Traditional legal pricing structure that became popular in the 1950s
• Per-Document Pricing - Alternative pricing model that aligns incentives between law firms and clients
• Context Engineering - Approach to providing comprehensive information to both human lawyers and AI systems
• Per-Customer Fine-tuning - AI training methodology focused on individual client or lawyer preferences rather than general consensus

Timestamp: [8:06-15:55]

Foundation Model Limitations and Custom Solutions

The 90% Accuracy Trap:

1. Foundation models excel initially - OpenAI, Anthropic, and Google provide excellent starting tools that reach 90% accuracy almost effortlessly
2. The final 10% is exponentially harder - Getting from 90% to 99% or 99.99% accuracy requires significant additional investment and customization
3. Enterprise requirements demand perfection - High ACV products need multiple nines of accuracy to deliver true five-star experiences

Custom Model Strategies:

• Per-customer prompt optimization - Adjusting prompts specifically for each client's needs and context
• Fine-tuning for enterprise data - Creating specialized models that understand client-specific contract language and requirements
• Access to proprietary data - Working with individual enterprise data that foundation model companies will never access

The Distribution Challenge:

• In-distribution data - Foundation models continue improving rapidly on data they've seen before
• Out-of-distribution advantage - Enterprise contract data remains hidden and proprietary, creating opportunities for specialized fine-tuning
• Competitive moat through exclusivity - Access to individual client data enables creation of highly customized RL fine-tuned models

Timestamp: [16:01-17:33]

The Core Value Proposition: Speed Over Everything

Primary Customer Benefits:

1. Unprecedented turnaround speed - Median contract completion times under one hour versus traditional weeks-long processes
2. Accelerated sales cycles - AEs and salespeople can respond to clients immediately instead of waiting for legal review
3. Strategic negotiation efficiency - Predicting optimal pushback points to eliminate unnecessary back-and-forth rounds

The Modern Business Reality:

• Everything is accelerating - Startup sales motions, hiring processes, and business operations are moving faster than ever
• Contract negotiations as bottleneck - The "API for business" has remained unchanged for 40 years since word processors
• Customer examples - Companies like Clay and Cursor benefit from this speed advantage in their own fast-moving operations

Smart Negotiation Strategy:

• Predictive term analysis - AI identifies which terms to accept and which three to push back on
• Turn reduction methodology - Strategic concessions save entire weeks by eliminating one negotiation round
• Quality threshold maintenance - Speed never compromises the essential legal protections clients need

Timestamp: [17:39-19:04]

The Unique Marriage of Technical and Legal Expertise

The Evaluation Challenge:

1. Engineers want quantitative metrics - John constantly requested formal evaluations and benchmarks
2. Lawyers provide instant qualitative assessment - Ryan could evaluate contract quality in 10 seconds through trained intuition
3. Visceral quality recognition - Lawyers develop taste-based judgment through training and risk profile understanding

The Collaboration Breakthrough:

• Initial scaling struggles - Hiring more engineers and lawyers separately felt like "revving the engine without clicking into gear"
• Integration catalyst - When lawyers received direct access to prompting tools and AI inputs/outputs, velocity increased dramatically
• Directional guidance - Lawyers provide immediate feedback on AI outputs, steering the technology in the right direction

Structural Advantages:

Legal Oversight Benefits:

• Insurance and liability coverage - Malpractice insurance and full liability assumption for all work
• Quality assurance - Expert validation ensures professional standards are maintained
• Escalation protocols - Knowing exactly when to involve human lawyers for complex decisions

Business Model Requirements:

• Liability as competitive advantage - Taking full responsibility for work quality is essential for business viability
• Certainty threshold - Must be completely confident in quality to stand behind the work professionally

Timestamp: [19:15-21:44]

Foundation Models and Custom Infrastructure

The Data Challenge:

1. State-of-the-art models lack contract specialization - Current foundation models aren't specifically tuned for contract work
2. Limited training corpus - Insufficient contract data exists in public datasets for proper model training
3. Data diversity bottleneck - Even within the legal domain, contract variety remains a significant challenge

Available Data Sources:

• Edgar SEC filings - The best publicly available dataset, but heavily overutilized and not applicable to smaller companies
• Hidden proprietary contracts - The most valuable contract data remains locked away in private company files
• Outsourced labeling limitations - Buying labeled datasets from large companies lacks the nuanced understanding needed

Technical Infrastructure:

Benchmarking and Evaluation:

• Continuous agent testing - Constant benchmarking across different environments to understand model strengths and weaknesses
• Quality score integration - Systematic measurement of performance tied back to legal quality standards
• Domain expert collaboration - In-house legal expertise provides competitive advantage over external data sources

Human-AI Collaboration Model:

• Lawyer-driven process - Big law experienced attorneys guide the AI systems and make final decisions
• AI summarization strength - Technology excels at explaining changes and predicting appropriate comments
• Language nuance challenges - AI struggles with subtle legal language differences that lawyers immediately recognize

Timestamp: [21:49-23:54]

The Subtlety of Legal Language

Critical Language Distinctions:

1. "Commercially reasonable" vs "reasonable" - These terms appear nearly identical in embedding space but carry substantially different legal meanings
2. Embedding space limitations - AI models struggle to distinguish between legally significant but linguistically similar terms
3. Context-dependent interpretation - Legal meaning often depends on broader contract context rather than individual word choice

AI Strengths in Legal Work:

• Summarization excellence - AI effectively explains contract changes and predicts appropriate comments for modifications
• Pattern recognition - Technology identifies common contract structures and standard clause variations
• Speed in routine tasks - Automated processing of standard terms and conditions

Human Oversight Requirements:

• Lawyer quality control - Experienced attorneys intervene on actual contract edits to ensure legal accuracy
• Nuanced decision making - Complex legal judgments still require human expertise and professional liability
• Client-specific risk assessment - Understanding individual client risk profiles and business contexts

Timestamp: [23:20-23:54]

Essential Insights:

1. The 90% accuracy trap - Foundation models reach 90% accuracy easily, but achieving 99%+ requires significant per-customer customization and fine-tuning
2. Deal velocity as competitive advantage - Crosby's under-one-hour contract turnaround times unlock speed for fast-moving startups in an unchanged 40-year-old process
3. Human-AI collaboration necessity - Engineers need lawyers' visceral quality assessment, while lawyers need AI tools to achieve unprecedented speed

Actionable Insights:

• Foundation model companies will never access proprietary enterprise contract data, creating opportunities for specialized fine-tuning
• Strategic negotiation (accepting some terms while pushing back on three key points) can eliminate entire weeks from deal cycles
• Vertical AI startups gain competitive edges by having domain experts in-house rather than outsourcing data labeling
• Taking full liability and malpractice insurance coverage is essential for AI legal services to gain enterprise trust
• AI excels at summarization and explanation but struggles with subtle legal language distinctions that lawyers immediately recognize

Timestamp: [16:01-23:54]

People Mentioned:

• Clay team - Customer example showcasing how Crosby's speed benefits fast-moving sales organizations
• Cursor team - Another discerning customer that values Crosby's rapid contract turnaround capabilities

Companies & Products:

• OpenAI - Foundation model partner providing excellent starting tools for AI development
• Anthropic - Foundation model partner contributing to Crosby's AI infrastructure
• Google - Foundation model partner in Crosby's technology stack
• Clay - Customer company that enables faster sales team operations
• Cursor - Customer company benefiting from Crosby's contract acceleration services

Technologies & Tools:

• Edgar SEC filings - Primary public dataset for contract training data, though overutilized and limited in scope
• Embedding space - Technical concept explaining why AI struggles with legally distinct but linguistically similar terms
• RL fine-tuned models - Reinforcement learning approach for creating customer-specific contract understanding

Concepts & Frameworks:

• Deal velocity - Core value proposition focusing on contract negotiation speed over traditional billable hour models
• The API for business - Crosby's metaphor for contract negotiations as the critical interface between companies
• Multiple nines of accuracy - Quality standard requiring 99% or 99.99% accuracy for enterprise applications
• Turn reduction methodology - Strategic approach to minimize back-and-forth negotiation rounds

Timestamp: [16:01-23:54]

AI-Powered Contract Communication

Crosby has discovered that AI's ability to explain legal reasoning is just as valuable as the legal work itself. When AI can articulate why certain contract language is being rejected or why specific terms are being pushed for, counterparties understand the position better and accept changes more readily.

Key Benefits:

• Reduced negotiation turns - Clear explanations lead to faster acceptance
• Better counterparty understanding - AI provides context for legal positions
• Faster resolution - Less back-and-forth when reasoning is transparent

Company-Specific Intelligence:

The AI system goes beyond generic legal knowledge to understand:

• Client business fundamentals - How tools like Cursor IDE or Clay work with sales data
• Industry context - What procurement teams need to know about specific purchases
• Contract implications - How business operations translate to legal terms

This contextual understanding allows lawyers to receive relevant data at the right time in the right parts of contracts, creating significant speed improvements while maintaining the natural feel of working with a knowledgeable legal advisor.

Timestamp: [24:00-24:48]

Seamless Legal Workflow Integration

Crosby has made a strategic bet on Slack as their primary interface, eliminating the need for separate legal software platforms. The process is designed to feel like consulting with a human colleague rather than using a formal legal tool.

Simple Process:

1. Tag @Crosby in Slack - No special interface required
2. Upload document - Send contracts directly through the platform
3. Receive results - Get reviewed documents back within hours with comments and explanations

Design Philosophy:

• No separate interface - Everything happens in existing workflow tools
• Natural interaction - Feels like talking to someone reviewing documents for you
• Email alternative - Also supports email workflow for different preferences

The integration removes friction from legal reviews by embedding the service directly into communication tools teams already use daily, making legal consultation as simple as asking a colleague for help.

Timestamp: [24:48-25:08]

Specialized Agent Infrastructure

Crosby has built a hierarchical system of AI agents that mirrors traditional law firm structure, with each agent optimized for specific contract types and tasks.

Current Implementation:

• Parallegal Agent - Routes incoming work from customers to appropriate human lawyers
• Specialized Contract Agents - Dedicated agents for NDAs, MSAs, DPAs, and other contract types
• Task-Specific Training - Each agent receives focused training for their particular specialty

Future Development Path:

The roadmap follows traditional law firm hierarchy:

1. Junior Associate Level - More complex contract analysis and drafting
2. Senior Associate Level - Advanced negotiation strategy and complex deal structures
3. Junior Partner Level - Strategic oversight and client relationship management

Success Factors:

• Quality Tools - Agents need sophisticated legal analysis capabilities
• Rich Context - Deep understanding of client business and legal history
• Focused Training - Specialization rather than generalization for better performance

This approach ensures agents can handle contract reviews quickly while maintaining the quality standards expected from experienced legal professionals.

Timestamp: [25:13-26:12]

Model Performance in Legal Tasks

The team has identified specific models that excel at legal reasoning and contract analysis through extensive testing and implementation.

Top Performing Models:

• Claude 3.5 Sonnet - Excellent thinking and reasoning capabilities for complex legal tasks
• Gemini 2.5 Pro - Strong performance specifically on legal document analysis
• Team favorite status - Claude 3.5 Sonnet has become so preferred that teammates create memes about the preference

Technical Approach Philosophy:

Rather than focusing heavily on fine-tuning for individual firms, the emphasis is on:

• Context engineering - Providing rich, relevant information to models
• Reinforcement learning techniques - Using reinforcement fine-tuning for comment generation and other specialized tasks
• Human-AI collaboration - Letting lawyers write prompts and provide domain expertise

The strategy prioritizes giving models better context and tools rather than extensive customization, allowing for more scalable and effective legal AI implementation.

Timestamp: [26:12-27:48]

The Context Engineering Advantage

Great legal AI success comes from replicating what exceptional product and commercial counsel do naturally - maintaining vast amounts of relevant information in working memory during contract negotiations.

Critical Context Elements:

• Company operations - Deep understanding of how the business actually works
• Negotiation history - Background on previous deals with specific counterparties
• Playbook knowledge - Familiarity with company-specific legal strategies and preferences
• Relationship context - Understanding of ongoing business relationships and priorities

Technical Implementation:

• Working memory loading - Getting all relevant context into the AI's active consideration
• Reinforcement learning - Using RL techniques for specialized tasks like comment generation
• Human-AI collaboration - Lawyers writing prompts and teaching domain knowledge

Psychological Safety Factor:

The approach maintains human oversight not just for quality, but because clients need the psychological comfort of knowing experienced lawyers are involved in their critical business negotiations.

Timestamp: [27:02-27:48]

The Magical Moment of Legal AI Adoption

Lawyers at Crosby experience a transformative moment when they first see AI handling tasks they've been doing manually, creating genuine excitement about the technology's potential.

Initial Lawyer Attitudes:

• Curiosity without experience - Many lawyers express interest in AI but haven't actually used it
• Skepticism about implementation - Questions about how AI will actually work in practice
• Eagerness to learn - Strong desire to be part of the AI transformation

The Breakthrough Experience:

One lawyer's reaction was particularly memorable - when she first saw the AI tools in action, the lights literally turned on in her expression. The realization that AI could handle tedious, repetitive work she didn't enjoy was described as "the most magical experience ever."

Weekly Transformations:

The team witnesses these breakthrough moments on a weekly basis as lawyers discover they can:

• Automate repetitive tasks - Let AI handle work they find tedious
• Focus on higher-value work - Spend time on strategic thinking rather than routine processing
• Teach AI their expertise - Transfer their knowledge to systems that can scale it

This emotional connection to AI capabilities drives adoption and innovation within the legal team.

Timestamp: [27:48-29:07]

Building Applied Legal Research Culture

Crosby has developed a unique culture that encourages lawyers to become AI prompt engineers and process innovators, moving beyond traditional billable hour metrics.

Cultural Foundation:

• Applied Legal Research Role - Following Harvey's lead in creating job titles that recognize AI-enhanced legal work
• Meta-awareness Praise - Public recognition for lawyers who step back and analyze their own work processes
• Process Innovation Rewards - Celebrating systematic thinking about legal workflows

Success Story Example:

One lawyer identified improvement opportunities and, after learning Miro, created a comprehensive process map so detailed and valuable that:

• It required professional printing due to its size and complexity
• Engineers were genuinely excited about the legal process documentation
• It became a model for cross-functional collaboration

Incentive Alignment:

• Speed-focused metrics - Lawyers are incentivized to reduce total review time rather than maximize billable hours
• Collaborative outcomes - Both lawyers and engineers work toward the same efficiency goals
• Innovation recognition - Public praise for finding better ways to handle repetitive work

Organizational Design:

The structure intentionally mixes lawyers and engineers on the same teams, requiring collaboration and shared accountability for results rather than traditional siloed legal work.

Timestamp: [29:07-30:50]

New York's AI-Ready Tech Evolution

Building an AI company in New York isn't contrarian when you understand the city's unique technical foundation and talent pipeline that perfectly aligns with AI development needs.

New York Tech Evolution:

1. First Wave (Early 2000s) - Ad tech explosion after the dot-com boom with companies like AppNexus
2. Second Wave - Technical talent from ad tech companies created dev tools like MongoDB
3. Parallel Development - Trading firms (Jane Street, HRT, Citadel) built sophisticated technical capabilities

Current Advantage:

• Senior engineering talent - Experienced engineers from both ad tech and trading backgrounds
• Growth expertise - Leaders who understand scaling from zero-to-one and one-to-ten
• Technical sophistication - Deep experience with complex, high-performance systems
• AI readiness - The combination of technical depth and business scaling experience

Strategic Positioning:

The convergence of ad tech algorithmic experience, trading firm technical rigor, and startup scaling knowledge creates an ideal environment for building sophisticated AI applications that need to operate at enterprise scale.

New York's tech ecosystem has evolved to the point where it offers unique advantages for AI companies, particularly those requiring both technical sophistication and business scaling expertise.

Timestamp: [30:50-31:53]

Essential Insights:

1. AI explanation capabilities - Teaching AI to articulate legal reasoning reduces negotiation turns and improves counterparty acceptance
2. Workflow integration strategy - Embedding legal AI directly in Slack eliminates interface friction and feels like consulting a colleague
3. Hierarchical agent architecture - Building specialized agents for different contract types and legal roles mirrors traditional law firm structure

Actionable Insights:

• Context engineering over fine-tuning - Focus on providing rich, relevant information rather than extensive model customization
• Culture transformation approach - Create incentives for lawyers to become AI innovators by rewarding process improvement over billable hours
• Geographic advantage recognition - New York's combination of ad tech, trading, and startup talent creates ideal conditions for sophisticated AI development

Timestamp: [24:00-31:53]

People Mentioned:

• Harvey (Legal AI Company) - Credited with creating the "applied legal research" job title and building center of gravity for lawyer-AI collaboration

Companies & Products:

• Cursor IDE - Code editor mentioned as example of client product requiring business context understanding
• Clay - Sales data platform referenced as example of client requiring specialized knowledge
• Slack - Primary integration platform for Crosby's legal services
• Miro - Process mapping tool used by lawyer for workflow documentation
• AppNexus - Ad tech company representing New York's first tech wave
• MongoDB - Database company example of New York's second tech wave
• Jane Street - Trading firm contributing to New York's technical talent pool
• HRT (Hudson River Trading) - Trading firm mentioned as part of New York's technical foundation
• Citadel - Trading firm contributing to New York's technical ecosystem

Technologies & Tools:

• Claude 3.5 Sonnet - Preferred AI model for legal reasoning and thinking tasks
• Gemini 2.5 Pro - AI model performing well on legal document analysis
• Reinforcement Learning - Technique used for comment generation and legal task optimization

Concepts & Frameworks:

• Applied Legal Research - Job title created by Harvey representing lawyer-AI collaboration role
• Context Engineering - Approach focusing on providing rich information to AI models rather than extensive fine-tuning
• Working Memory Loading - Technique for getting relevant context into AI's active consideration during legal tasks

Timestamp: [24:00-31:53]

New York's Evolution as a Startup Ecosystem

The Transformation Story:

1. Historical Context - In 2013, there might have been only one YC company in New York, showing how far the ecosystem has evolved
2. Talent Magnet Effect - Developers now specifically filter job searches by New York City, indicating the city's growing appeal
3. Breeding Ground Companies - Established firms like AppNexus and RAMP have become talent incubators, spawning new generations of startups

Key Advantages of Building in New York:

• Domain Expertise Density: Deep subject matter expertise in finance, creative fields, healthcare, and law
• Escape from AI Echo Chamber: Allows for more first-principles thinking and practical application
• Unique Development Process: "Dream in SF, build in New York" - distilling complex AI concepts into practical solutions

The RAMP Success Story:

A talented developer once applied to Parabus (early RAMP) for one simple reason: "I filtered by New York City on YC and you were the only company there." This anecdote illustrates how far New York's startup scene has progressed since then.

Timestamp: [32:00-33:33]

Building a Founder-First Culture

The Founder Factory Mentality:

1. Stepping Stone Philosophy - Crosby positions itself as a 4-5 year stepping stone for future founders
2. Autonomous Pod Structure - Engineers work directly with lawyers as customers, creating unique product development cycles
3. Founder Dinner Strategy - Leadership sends team members to founder events instead of attending themselves

New York-Specific Cultural Elements:

• High Bias Toward Entrepreneurship: More people wanting to start companies compared to other markets
• Design Emphasis: Strong focus on product design and taste, influenced by NYC's creative agencies
• RAMP's Cultural Impact: The success of RAMP has created a cultural shift throughout New York's startup ecosystem

Team Composition Strategy:

Founding Engineering Team: All members are either previous founders or aspire to be founders, bringing entrepreneurial mindset to product development

Unique Advantages:

• Direct Customer Feedback: Engineers work alongside lawyers who are their actual customers
• Rapid Product Iteration: Immediate feedback loops accelerate development cycles
• Authentic New York Elements: Even down to celebrating with real New York pizza (not Domino's)

Timestamp: [33:33-37:31]

The Mythology of Crosby

Multiple Origin Stories:

1. The Romantic Version - Founders took long existential walks through SoHo, repeatedly finding themselves on Crosby Street
2. The Symbolism - Crosby Street represents their vision: a beautiful old street in a modern neighborhood, combining traditional craftsmanship with modern innovation
3. The Practical Alternatives - Could be named after the hockey player, a hotel, or even their office manager's dog

What the Name Represents:

• New York Identity: Essentially a New York company at its core
• Blend of Old and New: Traditional artisanship combined with modern steel and glass architecture
• Multiple Interpretations: The ambiguity allows for various meanings and connections

The founders embrace the mystery and multiple interpretations of their company name, reflecting their approach to building something that honors both tradition and innovation.

Timestamp: [35:02-35:39]

The Future of Legal Automation

Market Structure Reality:

• Top Tier: 11,000 law firms (8% of total) generate 70-75% of industry revenue
• Long Tail: Remaining 92% of firms serve individuals with basic legal needs
• Underserved Market: The majority focuses on child support, leases, and personal legal matters

Complete Automation Prediction:

Individual Legal Services will be entirely automated because:

1. Current Alternative is Nothing - Many people receive no legal help today
2. Already Happening - People negotiate landlord leases using ChatGPT right now
3. Net New Market Creation - This isn't replacing lawyers; it's serving previously unserved populations

What Will Remain Human:

Corporate Law Firms and their jobs are quite safe for the foreseeable future due to complexity and stakes involved.

The Optimistic Vision:

Rather than replacing lawyers, AI will provide massive leverage - enabling lawyers to review 500 contracts per hour instead of 2, ultimately serving more people who currently have no legal access.

Mission: Building better legal infrastructure with technology to democratize legal services.

Timestamp: [37:31-39:32]

Total Turnaround Time as the Ultimate Measure

The Metric Definition:

Total Turnaround Time = Time from initial contract submission to final completion across all back-and-forth negotiations

Why This Metric Matters:

1. End-to-End Measurement - Captures the complete customer experience, not just individual touchpoints
2. Includes All Interactions - Accounts for contracts that require 2 back-and-forths versus those needing 5
3. Cumulative Time Tracking - Measures total time Crosby spends on each contract throughout the entire negotiation process

Operational Focus:

This metric drives the company's obsession with speed and efficiency, aligning with their goal of achieving contract turnaround times under an hour while maintaining quality legal work.

The metric reflects their fundamental shift from traditional billable hours to outcome-based pricing and service delivery.

Timestamp: [39:38-39:59]

Essential Insights:

1. New York's Startup Evolution - The city has transformed from having potentially one YC company in 2013 to becoming a major startup hub with dense domain expertise
2. Founder-First Culture - Crosby builds a culture where team members are treated as future founders, creating high autonomy and entrepreneurial mindset
3. Legal Automation Strategy - Individual legal services will be completely automated, while corporate law remains human-driven

Actionable Insights:

• Location Strategy: Consider New York for startups needing deep domain expertise in finance, law, healthcare, or creative fields
• Team Building: Hire people with founder aspirations and give them ownership of specific problem areas
• Market Opportunity: Focus on underserved legal markets where the alternative is currently "nothing"
• Success Metrics: Measure end-to-end customer experience rather than internal efficiency metrics

Timestamp: [32:00-39:59]

People Mentioned:

• Kareem - Shared the origin story of RAMP's early days when they were called Parabus
• Pat - Mentioned in context of a survey about where new graduates wanted to work

Companies & Products:

• RAMP - Financial technology company that evolved from Parabus, now considered a founder factory and design leader
• Parabus - Early name for what became RAMP, one of the first YC companies in New York (2013)
• AppNexus - Mentioned as one of the breeding ground companies for startup talent in New York
• Y Combinator - Referenced in context of filtering companies by location
• Neo - Mentioned as potentially conducting a survey about where new graduates wanted to work
• ChatGPT - Referenced as a tool people already use for legal advice and lease negotiations
• Sequoia Capital - Referenced as having portfolio companies in New York

Concepts & Frameworks:

• Founder Factory Model - The concept of companies serving as stepping stones for future entrepreneurs
• Pod Structure - Organizational approach where engineers work directly with domain experts (lawyers) as customers
• Total Turnaround Time - North Star metric measuring end-to-end contract completion time

Timestamp: [32:00-39:59]

Balancing Speed with Quality in Legal Negotiations

Core Challenge:

The fundamental tension in AI-powered legal services is maximizing deal velocity without compromising client negotiating positions. Traditional law firms optimize for billable hours, while Crosby optimizes for faster turnaround times - but speed alone could lead to poor outcomes for clients.

Crosby's Solution Framework:

1. Guardrail Metrics System - Dedicated team of lawyers and engineers focused exclusively on quality assurance
2. Customer-Aligned Quality Checks - Ensuring work meets both objective legal standards and client-specific risk profiles
3. Dual Metric Approach - Combining Total Turnaround Time (TAT) with quality guardrails

Key Metrics:

• TAT (Total Turnaround Time) - Primary speed metric that the entire office rallies around
• HURT (Human Review Time) - Measures time spent on human oversight and corrections
• Quality Guardrails - Ensures alignment with customer negotiation interests and risk tolerance

Implementation Strategy:

The company structures incentives around both speed and quality, with specific team members dedicated to preventing shortcuts that could harm client outcomes. This creates a system where faster delivery must still meet rigorous quality standards aligned with each client's specific negotiation strategy.

Timestamp: [40:40-42:20]

The Future of Agent-to-Agent Contract Negotiations

Current State vs. Future Vision:

Currently, Crosby's agents haven't negotiated against other AI agents in real-world scenarios. However, the founders envision a future where both parties deploy AI agents that can simulate entire negotiations before human involvement.

The Collaborative Negotiation Model:

1. Preference Capture - Each party's agent learns their specific risk tolerance, bottom lines, and negotiation style
2. Simulation Process - Agents run multiple negotiation scenarios to find optimal outcomes
3. Auditable Records - Complete documentation of negotiation progression from first offer to final terms
4. Human Oversight - Agents present simulated results for human approval before execution

Technical Implementation:

• Data Partitioning - Strict separation between opposing agents' access to confidential information
• Guardrail Systems - Agents only communicate through predetermined negotiation protocols
• Compute Allocation - More processing power dedicated to agents representing clients with higher stakes or complexity

Fundamental Philosophy:

Rather than adversarial negotiations, this approach treats contract discussions as collaborative problem-solving - finding mutually beneficial agreements faster through AI simulation and optimization.

Timestamp: [44:21-45:46]

Transformation of Legal Services and Career Paths

Industry Growth Patterns:

• 2007-2017: In-house legal teams grew 200% while law firms grew only 30%
• 2014-2024: In-house teams expanded from 320,000 to 440,000 professionals
• Driving Force: Legal work complexity requiring larger, more specialized in-house teams

Structural Advantages of In-House Teams:

1. Staffing Flexibility - Can hire lawyers, parallegals, legal operations specialists, and other roles
2. Creative Solutions - Fewer constraints than traditional law firm structures
3. Specialization - Ability to develop highly focused expertise areas
4. Cost Efficiency - More control over resource allocation and technology adoption

Future Predictions:

• AI-First Legal Companies - Specialized firms built around AI capabilities from the ground up
• Senior Partner Leverage - Experienced lawyers managing teams of AI agents rather than junior associates
• Agentic Systems - AI handling intake, document drafting, and review while humans focus on court appearances and strategy
• Golden Age of Innovation - Unprecedented transformation in legal service delivery methods

Career Impact:

The shift raises questions about opportunities for junior associates and paralegals, as AI may automate many traditional entry-level legal tasks while amplifying the capabilities of senior professionals.

Timestamp: [45:52-47:36]

Balancing Tradition with Innovation in Legal Education

Core Philosophy for Law Students:

Question everything while respecting brilliance - Challenge established practices and dogma while recognizing the fundamental value of legal academia and institutions.

Specific Recommendations:

1. Critical Thinking Approach

• Question traditional methods (even footnote formatting requirements)
• Challenge accepted "facts" that may be outdated or unnecessary
• Recognize that many established practices can be changed

2. Respect for Legal Foundation

• Understand that legal systems are what make society function
• Learn from brilliant legal academics and mentors
• Avoid arrogance while maintaining healthy skepticism

3. Practical AI Skills Development

• Learn how to effectively prompt and work with AI systems
• Understand how to harness AI power for legal applications
• Gain hands-on experience through apprenticeships with AI-first legal companies

Industry Reality Check:

The legal profession shows significant resistance to change, with some Fortune 500 general counsels refusing to try AI tools despite CEO encouragement. This creates opportunities for law students who embrace both traditional legal knowledge and modern AI capabilities.

Key Insight:

There's tremendous room for innovation in legal services, with many accepted practices being "permeable" and changeable - but this is often difficult to see from within traditional legal education structures.

Timestamp: [47:36-49:20]

Discovering the Human Element in Contract Negotiations

Research Context:

During the summer while researching the Crosby concept, Ryan spent time in India studying offshore legal services that handle routine contract negotiations for Fortune 500 companies.

Standard Offshore Process:

• Rule-Based Approach - Apply five standard rules to contract negotiations
• Mechanical Execution - Simple back-and-forth without creativity
• Wall-Like Negotiations - Predictable, inflexible responses that lack human insight

The Breakthrough Moment:

Ryan encountered one offshore legal worker who had developed an unconventional approach:

• Creative Solution - When negotiations got stuck, he would call the counterparty directly
• Human Psychology - Emphasized the long-term relationship and mutual benefits
• Unauthorized Innovation - He wasn't officially allowed to make calls but did it because "it works"

Key Realization:

This experience highlighted the essential human element in contract negotiations - they're fundamentally "an abstraction of a human-to-human conversation about what can we agree on."

Impact on Crosby's Philosophy:

The story reinforced their belief that while AI can dramatically improve efficiency and reduce transaction time, there remains a crucial role for human interface and "meeting of the minds" to reach true agreement between parties.

Timestamp: [42:48-44:15]

Essential Insights:

1. Incentive Alignment Challenge - Crosby solves the tension between speed and quality through guardrail metrics, ensuring faster delivery doesn't compromise client negotiating positions
2. Agent-to-Agent Future - The next evolution involves AI agents simulating entire negotiations between parties, creating auditable records and collaborative problem-solving approaches
3. Industry Transformation - In-house legal teams are growing 200% faster than traditional law firms, creating opportunities for AI-first legal service companies

Actionable Insights:

• Law students should question everything while respecting legal foundations, learning AI skills through hands-on apprenticeships
• Legal professionals must balance innovation with established expertise, as resistance to change creates competitive advantages for early AI adopters
• Contract negotiations retain essential human elements despite AI automation, requiring "meeting of the minds" for true agreement

Key Metrics and Concepts:

• TAT (Total Turnaround Time) - Primary speed metric for legal service delivery
• HURT (Human Review Time) - Measures human oversight requirements in AI-assisted legal work
• Collaborative Negotiations - Treating contract discussions as problem-solving rather than adversarial processes

Timestamp: [40:06-49:46]

People Mentioned:

• Stanford Law Professors - Described as brilliant mentors and some of the smartest people, representing the value of legal academia despite need for innovation
• Fortune 500 General Counsel (Telecoms) - Anonymous executive who refuses to try AI tools despite CEO encouragement, illustrating industry resistance to change

Companies & Products:

• Ramp - Financial technology company referenced as source of inspiration for creating memorable, "catchy" business metrics
• Fortune 500 Companies - Large corporations using offshore legal services for routine contract negotiations

Concepts & Frameworks:

• TAT (Total Turnaround Time) - Crosby's primary speed metric that the entire office rallies around for measuring legal service delivery
• HURT (Human Review Time) - Metric measuring time spent on human oversight and corrections in AI-assisted legal work
• Guardrail Metrics - Quality assurance systems ensuring AI work meets both objective legal standards and client-specific risk profiles
• Meeting of the Minds - Legal concept referring to mutual agreement and understanding between negotiating parties
• Data Partitioning - Technical approach to ensure opposing AI agents don't access each other's confidential information during negotiations

Industry Statistics:

• In-house Legal Growth - Teams expanded 200% from 2007-2017 while law firms grew only 30%
• Legal Professional Growth - In-house teams grew from 320,000 to 440,000 professionals between 2014-2024

Timestamp: [40:06-49:46]