20VC: Enterprises Will Not Adopt AI without Forward-Deployed...

The host and guest discuss the "chasm between model performance and adoption" in enterprise AI, explaining that despite improved models, businesses struggle with integration.

Fitzpatrick shares an anecdote about a bank CTO's rejection of an LLM tool due to data, security, and permission concerns, highlighting enterprise hurdles.

The discussion addresses whether enterprises are open for AI business, noting sector differences and the success rate of external vs. internal AI builds.

The conversation shifts to what CEOs should consider regarding CFO readiness for AI and the availability of appropriate talent.

Fitzpatrick debunks the myth that AI decision-makers must be highly technical, emphasizing the need for familiar business frameworks and outcome-based approaches.

The "Accenture paradigm" of layering many apps with extensive service contracts is contrasted with Invisible's approach to integrated AI solutions.

The advice to focus on 3-4 key initiatives instead of "science projects" is given, with the contact center identified as a common starting point.

The challenge of choosing between numerous AI vendors in the contact center space is explored, with a focus on differentiation and avoiding ineffective solutions.

The inaccuracy of out-of-the-box enterprise agents is highlighted, emphasizing the need for proof of concept and vendor selection based on demonstrated performance.

Invisible's strategy of offering free, eight-week proofs of concept to demonstrate tech efficacy is detailed.

The five modular components of Invisible's AI platform are explained, showcasing their adaptability across sectors.

The necessity of "forward-deployed engineers" for enterprise AI adoption is reiterated, emphasizing their role in custom workflow builds.

The economics of Forward Deployed Engineers (FDEs) are discussed, differentiating between solutions engineering and actual workflow configuration.

Invisible's FDEs typically complete customization in three months, requiring ongoing fine-tuning, unlike static SaaS solutions.

Invisible does not charge for FDEs, prioritizing proving their technology's effectiveness as a differentiator.

The advice for startups on adopting an FDE model is to consider it for workflow embedding and adoption changes, not just for repository-based systems.

The pricing model evolution is discussed, suggesting that integrated AI solutions will be more profitable due to faster customer acquisition and value delivery.

Invisible's business model is clarified as an AI training platform, not just a talent marketplace, emphasizing the complex process of sourcing and validating expertise.

Revenue concentration in the AI training space is acknowledged, with Invisible aiming for greater diversification through enterprise adoption.

The negotiation dynamic between customers and data providers is discussed, with a focus on willingness to pay for good, validated data.

The importance of human data and expertise in AI training, especially for complex, multimodal, and context-specific tasks, is emphasized.

The market for AI is seen as being in its "first inning," with significant growth potential for a decade as enterprises adopt and fine-tune AI.

The difficulty of scaling an organization that consistently delivers high-quality, specialized data within tight deadlines is highlighted.

The increasing specialization of data requirements and the acquisition of niche data pools are recognized as key market trends.

The role of pay and price discovery in the talent marketplace is discussed, emphasizing the importance of finding the right expert at a fair price.

The concept of controlling a finite supply of data providers is questioned, with the focus being on balancing supply and demand dynamically.

Switching barriers in data provision are considered, with Invisible's expertise in complex tasks like fine-tuning models for underwater drone swarms cited as an example of built-in value.

The importance of sector-specific expertise and developing logic around specialized tasks is emphasized for sustainable growth.

The shift from general public benchmarks to task-specific performance metrics for AI is discussed.

The debate on whether frequent model updates matter is addressed, with the focus shifting to hyper-specific performance and trust for enterprise adoption.

The challenge of enterprise AI adoption is linked to achieving 99% precision on specific tasks, requiring a process similar to model risk management in banking.

The potential impact of AI on junior roles and the future talent pipeline is discussed, suggesting that AI will augment, not eliminate, many jobs, leading to more sophisticated work.

The analogy of Excel replacing slide rules in accounting is used to illustrate how technology can increase complexity and the scope of work, rather than reducing headcount.

The future of AI in enterprises is projected to involve 3-5 key players specializing in different niches, rather than a single dominant force.

Palantir is identified as a respected company in enterprise AI for their early recognition of the importance of forward-deployed engineering and customization.

The distinction between revenue and bookings in the context of services and talent acquisition is clarified, with the argument that service-based revenue is genuine.

The misconception that synthetic data will replace human feedback is addressed, with emphasis on the ongoing need for human expertise in complex AI tasks.

The difference between traditional ML and GenAI is highlighted, with GenAI requiring human-in-the-loop processes for validation and fine-tuning.

Invisible Technologies is not profitable currently due to heavy investment in technology and growth, reflecting a strategic choice to prioritize scaling.

The decision to invest heavily in growth is driven by the current unprecedented environment and Invisible's unique positioning.

The company is not investing in physical world interactions due to the high cost and time required, despite recognizing their potential.

Invisible's brand building efforts have increased, moving from a low public profile to a more strategic approach focused on trust and consistent delivery.

The importance of aligning public and private narratives, and avoiding over-promising, is stressed for building brand trust.

The company differentiates itself by being transparent about the nuances of AI and avoiding claims of out-of-the-box solutions.

The non-deterministic nature of AI introduces risk, requiring careful delivery and validation, unlike traditional SaaS products.

Invisible focuses on AI agents and workflows, recognizing that many current "agents" are based on traditional scripting rather than true AI.

The discussion touches on the limitations of current robotics and the need for task-specific solutions.

The guest shares an experience of learning and iterating in the early days of AI, emphasizing the importance of customer trust and recruiting talent.

The core of building a successful company is identified as recruiting exceptional people and creating an environment where they can thrive.

The focus on recruiting great people, fostering a positive culture, and enabling them to build great things is highlighted.

Advice for startup CEOs on recruiting, retaining, and evolving talent emphasizes hiring adaptable "all-around athletes" and creating an enjoyable work environment.

A counterpoint is made that "culture is bullshit" and winning matters, but the guest clarifies this applies more to execution-focused businesses than research-driven AI companies.

The concept of "war mode" is considered in the context of AI research, with the acknowledgement that some delivery operations teams operate in that mode.

The biggest fear is the decision to pursue hyperscale growth, which requires significant capital investment, versus steady, consistent growth.

The company is investing in areas like enterprise and custom platforms, differing from peer AI training companies.

Invisible transitioned from a remote company to having physical offices in multiple cities, believing it strengthens culture, customer relationships, and productivity.

Productivity is seen to increase with in-person collaboration, especially for tackling complex engineering problems.

The balance between flexibility and in-person collaboration is discussed, with a preference for in-office presence on weekdays and flexibility on weekends.

Two key management beliefs that have changed are the fallacy of control in a complex environment and the overrated concept of strategy in the rapidly evolving AI world.

The rapid pace of technological change in AI necessitates constant adaptation and iteration, making five-year strategic plans less relevant than building interoperable frameworks.

Balancing travel with relationships requires a supportive partner and proactive communication, recognizing that the demanding travel schedule is not permanent.

The quick fire round touches on AI company valuations and the most underrated companies in the information space.

The best advice received is to focus on recruiting great people, creating a positive culture, and making them rich.

The widely held belief that out-of-the-box AI agents will solve everything is considered wrong, as AI requires training, fine-tuning, and process redesign.

The investment thesis is to bet on AI agents that are more like SaaS products and on business models that leverage AI, particularly in the physical world.

The potential for lower margins in the AI business is questioned, with the idea that integrated solutions might be more profitable due to faster customer acquisition.

The guest is most excited about AI's potential to positively impact energy efficiency, healthcare cost reduction, and education accessibility.

The shift in educational systems and talent assessment towards valuing skills and cognitive aptitude over traditional college degrees is seen as a positive development.