Loading summary...
6 min read (75% time saved)
An MIT report stating 95% of enterprise AI projects fail is often misinterpreted; the reality is that large companies struggle to build effective AI internally, creating a massive opportunity for startups that can deliver integrated, high-quality AI solutions by understanding business processes and navigating enterprise politics.
Summarize another video
Press ⌘K to quickly paste a new URL
The discussion opens by addressing the viral claim that 95% of enterprise AI projects fail, often used by "AI influencers" to dismiss AI as a scam. The panelists clarify that the MIT report, when properly understood, actually confirms the viability of AI agents and approaches that are working in the real world. This misinterpretation highlights a different go-to-market strategy for AI solutions beyond standard enterprise sales, emphasizing deep integration into business processes and systems of record.
Enterprises typically attempt AI adoption through internal IT or large consulting firms like Ernst & Young or Deloitte. However, internal IT systems are often subpar, and engaging consultants can introduce additional problems. The high failure rate is attributed to the generally poor quality of software built within large organizations. Even a company with infinite resources like Apple struggles to create a functional calendar app, illustrating the difficulty of building good software, let alone complex AI, within any large entity.
Deploying sophisticated software in large enterprises is challenging due to political battles and turf wars across multiple teams. Consultants are often brought in to mediate and create specifications, but they frequently lack the technical expertise to implement the solutions. Internal software teams face issues with old, siloed systems. The result is often a "camel" (a horse designed by committee) rather than an effective solution. Examples like Tactile, which builds business decision engines for banks, demonstrate how startups can deliver real-time, pluggable AI models for a fraction of the cost and time compared to banks' internal efforts. Another case, Greenlight, successfully implemented an AI system after a bank's year-long, failed attempt with Ernst & Young. The MIT report indicates that enterprises primarily try to build in-house (two-thirds of projects), but external vendors (one-third of projects) have a much higher success rate. This is largely due to the scarcity of "polymaths" who are skilled in both product and engineering, and can understand user needs, leaving a "startup-shaped hole" in the market.
Reduct, a company specializing in document processing for AI, secured a significant deal with a large "fan company" just 154 days after their YC batch. This fan company had previously failed to build an internal solution for years, trying various open-source tools and OCR solutions like AWS Tesseract. Reduct's product excellence allowed them to win the deal, despite having to navigate internal politics, as highlighted in the MIT paper. Their solution has been in production for over a year, demonstrating successful external AI implementation.
To succeed in selling to large enterprises, startups should identify and befriend internal champions. These champions are often individuals who secretly harbor startup dreams but are too risk-averse to pursue them. They can live vicariously through the startup's journey and are motivated to see the startup succeed, helping navigate internal politics. Startups should also maintain authenticity rather than trying to mimic the formality of large corporations.
Another effective tactic is to seek out founders whose companies were acquired by large enterprises. These individuals, now embedded within the larger company, can act as powerful advocates and guides. Examples include a YC company, Q, acquired by Apple, which helped Triple Bite work with Apple, and a founder who sold his company to Oracle, assisting with procurement and providing an internal playbook for a pilot program. This "pay it forward" culture is a unique aspect of Silicon Valley.
The MIT paper also reveals an overwhelming demand from enterprises for AI adoption and a greater willingness to bet on new startups. While enterprises would prefer to buy from established software companies, these companies often cannot build the necessary products. This inability stems from internal engineering teams that are skeptical of AI and codegen tools, hindering internal development. This creates an opportunity for startups to provide solutions that enterprises cannot build themselves. The panelists encourage engineers to overcome their skepticism and embrace AI tools, as they can significantly boost productivity. The discussion also touches on the misinterpretation of expert opinions, where some may selectively hear that AI is "overhyped" instead of recognizing the vast opportunities for building supporting tools and infrastructure.
The exciting takeaway is the immense opportunity to rebuild existing systems to be AI-native, creating numerous opportunities for founders. A key quote from an enterprise buyer highlights that once time is invested in training an AI system, switching costs become prohibitive, creating a strong "moat" for successful AI solutions. The video concludes by reiterating that while AI implementation is hard, successful startups, often from YC, demonstrate that highly skilled product people and engineers can achieve success, understanding both technology and human needs. This is good news for founders, as they can be part of the successful 5% if they are truly excellent.
"If your engineers don't believe in this, then how are you going to build a product that actually works?" — [Garry Tan]
"The output of the study is no surprise to me in that the majority of software that actually gets built in the world is very, very bad." — [Jared Friedman]
"Apple, a company with infinite resources and infinite access to the smartest people in the world, cannot make a good calendar app." — [Jared Friedman]
"For now there's just this startup-shaped hole in basically every process or every sort of annoying system that should exist that doesn't exist yet." — [Jared Friedman]
"There's overwhelming demand from enterprises to adopt AI and they're way more willing to take bets on new startups." — [Diana Hu]
"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." — [CIO of a $5 billion financial services firm (quoted)]
No Priors: AI, Machine Learning, Tech, & Startups
7 min read
No Priors: AI, Machine Learning, Tech, & Startups
8 min read
My First Million
11 min read
Freeplay
7 min read
Y Combinator
7 min read
No Priors: AI, Machine Learning, Tech, & Startups
7 min read
