Invest Like the Best with Patrick O'Shaughnessy

Jesse Zhang - Building Decagon - [Invest Like the Best, EP.443]

October 6, 2025

Key Takeaways Copied to clipboard!

  • Founders with backgrounds in highly competitive environments like math and coding contests often possess the problem-solving aptitude and competitive drive necessary for success in current high-stakes company building, especially in AI. 
  • The most effective early-stage customer discovery involves systematically asking potential customers highly specific, uncomfortable questions about exact willingness to pay and organizational ROI hurdles to gain genuine signal on product-market fit. 
  • Customer service is a leading enterprise AI use case because it offers easily quantifiable ROI (cost savings) and inherently supports safe, staged rollouts via existing human escalation paths, unlike riskier, non-deterministic AI applications. 
  • Founders should use the fundraising window as a critical test to gauge an investor's willingness to be helpful post-investment, as helpfulness during the pursuit is a strong proxy for future support. 
  • The best investors underwrite a business by deeply understanding its customers and assessing company culture, recognizing that talent attraction is a key indicator of future success. 
  • In the current AI landscape, companies should prioritize capturing market share and user mindshare over immediate gross margins, betting on the exponential decrease in model serving costs over time. 

Segments

Competitive Culture and Mindset
Copied to clipboard!
(00:05:04)
  • Key Takeaway: A culture emphasizing relentless focus, competition, and intensity is viewed as a durable competitive advantage when building in highly contested, high-growth technology markets.
  • Summary: Decagon’s office features a quote promoting overcoming challenges and defeating enemies, reflecting a competitive culture designed to win in the exciting but crowded AI industry. This mindset attracts founders and talent who grew up in similarly intense, objective-driven environments like math contests. Success in company building is often linked to strong problem-solving capabilities derived from this rigorous training.
Systematic Ideation and Discovery
Copied to clipboard!
(00:12:00)
  • Key Takeaway: Founders should systematize ideation by rigorously testing hypotheses through deep customer qualification questions focused on exact pricing and organizational ROI justification.
  • Summary: Jesse Zhang learned from a previous venture’s bumpy journey to systematize the ideation process for Decagon by focusing on strengths like rational problem-solving. Conversations must move beyond general interest to ask specific questions like, ‘Exactly how much would you pay for this?’ and ‘Who needs to approve the ROI?’ This process filters out weak ideas where customers offer vague praise but cannot commit financially.
Decagon’s AI Customer Service Focus
Copied to clipboard!
(00:20:21)
  • Key Takeaway: Customer service emerged as the conviction-backed focus because its ROI is easily quantifiable, and the existing infrastructure naturally supports safe, staged deployment of non-deterministic AI agents.
  • Summary: Decagon focuses on AI customer service agents because the ROI is simple to calculate by measuring potential deflection from existing human agent costs. Furthermore, customer service environments allow for low-risk deployment; the natural escalation path to a human agent provides comfort to enterprises adopting new, non-deterministic AI technology.
AI Use Case Spectrum Comparison
Copied to clipboard!
(00:23:42)
  • Key Takeaway: Successful AI agent adoption targets either augmenting the highest-paid talent (like engineers) or replacing the most replaceable, high-churn labor (like Tier 1 customer service).
  • Summary: AI use cases can be mapped across a spectrum based on the cost of the human labor being replaced. Coding agents augment highly paid engineers, leading to productivity gains that are easy to report. Conversely, customer service agents often replace outsourced, high-turnover labor where full automation is feasible and cost-saving is immediate.
Agent Deployment and Guardrails
Copied to clipboard!
(00:26:31)
  • Key Takeaway: Successful enterprise AI deployment requires establishing a quantifiable ‘what good looks like’ standard via simulation suites before going live, effectively creating a captive reinforcement learning process.
  • Summary: Before deployment, companies must align stakeholders on rigorous evaluation metrics covering tone, accuracy, and brand guidelines, which is often difficult as no single person knows all the answers. Decagon builds simulation suites to test performance against these standards, allowing for continuous, quantifiable improvement of the agent before it handles live traffic.
Voice AI and Latency Challenges
Copied to clipboard!
(00:32:10)
  • Key Takeaway: The next major frontier for customer experience is voice AI, but achieving human-like interaction requires voice-to-voice processing to capture tone and reduce latency, despite current high hallucination rates.
  • Summary: Voice is the most natural human UI, but the bar for AI voice is high due to the uncanny valley effect. Voice-to-voice models are superior because they capture cadence and tone, but they currently suffer from significantly higher hallucination rates (estimated at 8x higher than text). Enterprise workflows require low latency, which is difficult when agents must first gather data via APIs before responding.
Data Utilization and Agent Moats
Copied to clipboard!
(00:39:15)
  • Key Takeaway: The primary moat for AI agents is the continuous, automated improvement derived from ingesting and structuring unstructured customer interaction data over time, leading to superior performance versus new entrants.
  • Summary: Customer interaction data, historically unstructured and underutilized, can now be processed by LLMs to extract topics and flag areas needing improvement automatically. This continuous learning loop allows agents to evolve based on real customer conversations and internal SOPs, creating a performance gap that is difficult for competitors to close.
Future of Conversational UI
Copied to clipboard!
(00:44:08)
  • Key Takeaway: The ultimate vision for AI agents is to become the unified, personalized conversational UI for a brand, handling sales, support, and actions seamlessly, much like a digital concierge.
  • Summary: The end state involves a unified brand agent that serves as the primary interface for all user interactions, potentially replacing the need to visit a website or app. The best customer experience is enabled when the agent has deep context and the necessary internal APIs to take complex actions, such as processing a credit card replacement request end-to-end.
Investor Demand and Dynamics
Copied to clipboard!
(00:49:13)
  • Key Takeaway: Investor excitement in AI companies with real enterprise traction is currently high, leading to preemptive fundraising rounds.
  • Summary: There is significant investor excitement around AI companies demonstrating real traction, often resulting in preemptive funding offers. Founders should leverage this period of high investor desire to test potential partners on their helpfulness before committing them to the cap table. The best investors focus on the business’s current performance and future potential rather than just previous valuations.
Investor Selection Criteria
Copied to clipboard!
(00:51:45)
  • Key Takeaway: The best investors demonstrate their future utility by actively helping founders during the fundraising window and exhibit high raw intellectual throughput.
  • Summary: Investors who are unwilling to help until they are invested are difficult to assess for future usefulness; the best ones actively signal their network access and problem-solving capabilities early on. Founders should prioritize investors with high cognitive ability who think from first principles regarding the company’s unique challenges, especially in rapidly evolving fields like AI.
Underwriting Best Practices
Copied to clipboard!
(00:53:48)
  • Key Takeaway: Effective investor underwriting relies heavily on deep customer understanding and cultural alignment, as many customer references may be unreliable.
  • Summary: Top investors gain a deep understanding of the company by researching customers before initial calls, recognizing that some customer references may be inaccurate or misleading. They also index heavily on company culture, understanding that a place where good talent congregates is a strong positive signal for the business’s long-term health.
Decagon Culture and Talent Wars
Copied to clipboard!
(00:54:56)
  • Key Takeaway: Decagon’s culture emphasizes extreme competitiveness, bias to action, and attracting talent motivated by hard work, strong relationships, and career leapfrogging.
  • Summary: Winning highly sought-after AI talent requires the entire team to swarm candidates, understanding their personal career goals, and designing roles around them. The company attracts individuals who view this period as the prime of their career, expecting hard work in exchange for strong financial outcomes and lifelong professional relationships.
Proprietary Models vs. Fine-Tuning
Copied to clipboard!
(00:58:25)
  • Key Takeaway: The future involves a balance where smaller, fine-tuned models handle specific agent tasks, while large foundation models provide necessary general intelligence.
  • Summary: While foundational models from labs like OpenAI remain crucial for general intelligence, applications are increasingly using smaller, fine-tuned models for specific, mature agent functions to improve latency and performance. Fine-tuning was initially avoided due to rapid model changes, but it is now viable for specialized tasks within a larger agent architecture.
AI Capabilities Over/Underestimation
Copied to clipboard!
(01:03:33)
  • Key Takeaway: Enterprise adoption of AI is slower than public perception suggests due to the non-determinism requiring near-perfection, but exponential improvements in cost are underestimated.
  • Summary: Enterprise use cases lag because the bar for AI success is set near-perfectly, unlike human performance which is evaluated holistically against its own mistakes. Conversely, the public underestimates the exponential improvement curve in performance and cost reduction, meaning current high operational costs for applications like coding agents are temporary.
Application Layer Value and Margins
Copied to clipboard!
(01:07:22)
  • Key Takeaway: The application layer captures the most value by building extensive software tooling around models, which is harder for foundation model labs to replicate.
  • Summary: Decagon maintains healthy margins by operating in the application layer, where the value captured is highest because the software solves the core business problem, not just the underlying model cost. Building substantial software functionality—like observability, guardrails, and QA testing—creates defensibility against simple model swaps or updates from foundation model providers.
Qualifying Ideal Enterprise Customers
Copied to clipboard!
(01:11:40)
  • Key Takeaway: The best enterprise customers are led by intellectually curious leaders who are genuinely committed to rapid AI adoption and provide aggressive, high-quality feedback.
  • Summary: Ideal customers are those whose leadership is genuinely excited about AI and willing to cut through internal bureaucracy to implement solutions quickly. These partners provide substantial, data-centric feedback, contrasting sharply with those who only adopt AI due to a board mandate without deep commitment.
Startup Worldviews and Milestones
Copied to clipboard!
(01:13:36)
  • Key Takeaway: Jesse Zhang prioritizes commercial practicality and customer willingness-to-pay above purely intellectual or technical pursuits when building a company.
  • Summary: Jesse Zhang emphasizes a practical, commercial focus, believing that optimizing for the highest likelihood of success requires constantly asking customers how much they would pay for the solution. Team motivation is effectively driven by rallying around clear, visible milestones, sometimes symbolized by small rewards like custom jackets, which foster unity toward a common goal.
AI Hype vs. Reality
Copied to clipboard!
(01:16:16)
  • Key Takeaway: The current emphasis on ‘forward-deployed engineers’ is often misapplied in startups without the massive client base required for that high-cost, high-touch model.
  • Summary: The concept of ‘forward-deployed engineers’ (FDEs), popularized by firms like Palantir for multi-million dollar deals, is often inappropriately adopted by startups serving smaller clients, making scaling impossible. A business constrained entirely by the need to hire good people for bespoke work cannot scale quickly, necessitating a balance between hands-on customer support and scalable processes.