Good News For Startups: Enterprise Is Bad At AI

Key Concepts

AI Project Failure Rate: The perceived high failure rate of AI projects, often cited by influencers.
MIT Study: A specific study that has been misinterpreted by some to claim widespread AI project failure.
Enterprise AI Adoption: The challenges and strategies involved in large organizations implementing AI solutions.
Internal IT Systems: The often inadequate and siloed nature of existing IT infrastructure within enterprises.
Consulting Agencies: The role of firms like Ernst & Young and Deloitte in enterprise AI projects and their limitations.
Product Excellence: The critical factor for startups succeeding in enterprise AI sales.
Systems of Record: Deep integration into an enterprise's core data and operational systems.
Go-to-Market Strategy: The unique approach required for AI solutions, differing from traditional SaaS sales.
Polymaths: Individuals with a broad range of skills, essential for successful AI product development.
Startup-Shaped Hole: The gap in the market for startups to build solutions for processes that existing systems don't adequately address.
AI Native Systems: The need to rebuild software from the ground up to effectively leverage AI.
Switching Costs: The high cost and effort associated with changing systems once an AI solution is implemented, creating a "moat" for vendors.
"Doomer" Narrative: The tendency to focus on AI's limitations and failures rather than its potential.
"Things That Don't Scale": Early-stage tactics, like building personal relationships, crucial for startup success.
Champion: An internal advocate within an enterprise who champions a startup's solution.

Summary

This discussion debunks the common narrative that AI projects have a high failure rate, particularly as propagated by some AI influencers citing a misleading interpretation of an MIT study. The speakers argue that the study, when read thoroughly, actually confirms the effectiveness of certain AI approaches and highlights the unique challenges and opportunities in enterprise AI adoption.

The Misinterpretation of the MIT Study and the Reality of AI Projects

The prevalent narrative, amplified on platforms like X and YouTube, claims that 95% of AI projects fail, using this as evidence that AI is a scam. However, the speakers, particularly Jared, who dug into the MIT report, found that the tweets and summaries circulating were misleading. Instead of proving AI's failure, the study, in their view, validates many of the podcast's long-held beliefs about AI agents and successful implementation strategies. The core issue isn't that AI itself fails, but rather the difficulty enterprises face in implementing it effectively.

Challenges in Enterprise AI Adoption

Enterprises often struggle with AI adoption due to several interconnected factors:

Inadequate Internal IT Systems: Existing internal IT systems are frequently described as "bad," outdated, and siloed, making integration of new AI solutions difficult.
Limitations of Consulting Agencies: While enterprises may turn to consulting firms like Ernst & Young or Deloitte when internal IT fails, these firms often lack the deep technical expertise to build sophisticated software. Their role is more about mediation and documentation, leading to a "horse designed by a committee" outcome – a product that satisfies multiple stakeholders but is ultimately ineffective.
Political Battles and Turf Wars: Large enterprises are prone to internal politics and competing interests, which complicate the deployment of any new, cross-functional software. Consultants can act as mediators, but the implementation phase remains challenging.
Lack of Internal Expertise: The speakers posit that there aren't enough individuals within large organizations who possess the necessary blend of product vision and engineering prowess to build functional AI systems. Many engineers are "in the coding cave" and lack domain understanding, while domain experts may lack coding skills.

The Success of Startups in the AI Space

The difficulties faced by enterprises create a significant opportunity for startups. The speakers argue that if a startup can successfully build a working AI product, enterprises will be receptive because they have limited internal options and cannot easily turn to established companies that also struggle with AI implementation.

Examples of Startup Success:

Tactile: This company built a "business decision engine" for banks, handling KYC and AML processes in real-time. Major banks like Citibank and JP Morgan had attempted to build similar systems internally for years, costing tens of millions of dollars. Tactile, however, delivered a REST API solution that could integrate the latest AI models, all at a fraction of the cost and time.
Greenlight: This company sells AI systems to banks. They encountered a situation where a bank rejected their solution due to an existing relationship with Ernst & Young, who promised to build an AI system. After a year of failure by Ernst & Young, the bank returned to Greenlight, whose system was then successfully deployed.
Castle AI: This company develops an AI mortgage underwriter. They often face "bake-offs" against incumbent vendors who have been in the market for decades. While these vendors are now adding AI to their existing subpar systems, Castle AI's native AI approach, built with product taste from the beginning, has allowed them to win deals.
Reduct: This company specializes in document processing for AI. A large fan company approached Reduct after years of failed attempts to build internal solutions using open-source tools, AWS Textract, and other OCR solutions. Reduct's product excellence and ability to navigate internal politics led to a significant deal, with their system live in production for over a year.

The "Startup-Shaped Hole" and the Need for Polymaths

The current landscape presents a "startup-shaped hole" in many business processes. This gap exists because there's a rare combination of skills required: deep understanding of the latest AI advancements, strong product taste, and the ability to empathize with and integrate human processes into a product. These individuals are described as "polymaths" – individuals who are good at both product and engineering, and can bridge the gap between technical capabilities and business needs.

Strategies for Startups to Succeed in Enterprise Sales

The discussion highlights several strategies that enable startups to overcome enterprise adoption hurdles:

Deep Integration ("Groing into Systems of Record"): Unlike traditional SaaS which is often plug-and-play, successful AI solutions require deep integration into an enterprise's core systems. This is time-consuming but leads to significant rewards.
"Things That Don't Scale": Building personal relationships and finding internal "champions" is crucial. Reduct, for instance, became good friends with their champion, who saw them as "smart kids" and wanted to support them. This champion often embodies the entrepreneurial spirit that the enterprise employee might have suppressed.
Authenticity: Founders should be authentic and not try to mimic the formalism of large corporations (e.g., wearing suits). Being smart and competent is key, but copying corporate styles is unnecessary.
Leveraging Acquired Companies: Founders of companies acquired by large enterprises can act as valuable champions, providing insights into internal politics and procurement processes. Examples include Triple Bite working with Apple through a founder of a previously acquired YC company, and a founder who sold their company to Oracle helping Triple Bite secure a pilot.
Focus on Product Excellence: As seen with Reduct and Castle AI, superior product quality and a native AI approach are critical differentiators against incumbent, often superficial, AI solutions.

The Overwhelming Demand and the "Moat"

Despite the implementation challenges, there is "overwhelming demand" from enterprises for AI solutions. A key quote from a CIO of a $5 billion financial services firm highlights this: "We're currently evaluating five different gen AI solutions. But once we've invested time in training a system, the switching costs will become prohibitive." This quote is presented as direct evidence of a "moat" for AI vendors, countering the idea that AI solutions lack defensibility.

The Role of Engineers and the "Try It" Message

The speakers strongly encourage engineers to embrace AI tools like codegen. They argue that many engineers are resistant due to a belief that AI is overhyped, fueled by narratives like the MIT study. The message to these engineers is simple: "just try it." Investing time in learning and using these tools can transform 1x engineers into 10x engineers and 10x engineers into 100x engineers. The resistance is seen as an emotional barrier that needs to be overcome.

Rebuilding Systems to be AI Native

The ultimate opportunity lies in rebuilding existing software systems to be "AI native." This requires a complete rewrite to effectively leverage AI, presenting a vast landscape of opportunities for founders to create new, AI-centric solutions.

Conclusion

The prevailing narrative of AI project failure is largely a misinterpretation of implementation challenges within enterprises. The MIT study, when understood correctly, points to the difficulty of building AI solutions internally or through traditional consulting. This difficulty, however, creates a significant advantage for startups that can deliver genuinely effective, deeply integrated AI products. The key to success lies in product excellence, understanding enterprise needs, building strong internal relationships, and for engineers, overcoming skepticism and embracing AI tools. The demand for AI is high, and the switching costs for successful implementations create strong moats, indicating a bright future for startups that can navigate these complexities. The message is optimistic: for those who are truly good at technology and possess a polymathic skillset, the opportunity to be part of the successful 5% of AI implementations is very real.