How AI entrepreneurs are getting "The Lean Startup" wrong

Eric Ries 's Lean Startup methodology made such an impression on me during his time on my board at Code for America that I went on to teach it to grad students at the Harris School of Public Policy at the University of Chicago . Which makes me particularly bemused watching today's AI startups completely misinterpret its core principles.

The contradiction is striking. Ries advocated for using human "concierges" as temporary solutions to validate business hypotheses before investing in expensive automation. Yet today's AI startups are doing precisely the opposite – deploying expensive, general-purpose AI as a temporary solution without a clear path to economic sustainability. It's a fundamental misunderstanding of both technological progression and business model validation.

In the current gold rush of generative AI startups, a curious paradox has emerged. While many founders rush to build businesses atop large language models like GPT-4, a quieter but potentially more revolutionary trend is taking shape: the rise of small language models (SLMs). This shift isn't just about technology – it represents a fundamental rethinking of how AI startups should approach building sustainable businesses.

The Current Landscape: The Concierge Trap

Today's AI startup landscape is littered with what we might call "GenAI concierge" businesses. These startups follow a familiar pattern: they leverage powerful general-purpose AI models to automate services that traditionally required human expertise. It's an appealing pitch – replace expensive human labor with AI that can work 24/7 at a fraction of the cost.

But there's a critical flaw in this logic. Unlike traditional "Wizard of Oz" startups that used human labor as a temporary stand-in for eventual automation, these AI concierge startups are using expensive, general-purpose AI as a temporary solution for... what exactly? The unit economics often don't improve with scale, and API costs are more likely to increase than decrease over time, as we've already seen with recent price hikes from major providers.

The Small Language Model Opportunity

This is where small language models enter the picture. Instead of using heavyweight, general-purpose AI models for every task, forward-thinking startups are beginning to deploy smaller, specialized models that are:

• Trained on specific domains
• Optimized for particular tasks
• More cost-effective to run
• Faster and more reliable in their narrow focus
• Easier to control and customize

This approach isn't just more economically sustainable – it's often more effective. A small model trained specifically on medical terminology will likely outperform a general-purpose model on medical tasks, while consuming far fewer resources.

Economic Advantages of Going Small

The economic benefits of this approach are compelling:

1. Lower Operating Costs

• Reduced compute requirements
• Potential for on-premise deployment
• Less dependency on expensive API calls

2. Better Scaling Economics

• Costs grow more linearly with usage
• More predictable unit economics
• Easier capacity planning

3. Greater Business Control

• Less vulnerability to API pricing changes
• Better data privacy and security
• Ability to optimize for specific business metrics

Beyond Economics: The Strategic Advantage

The shift to smaller models isn't just about cost savings – it represents a fundamental strategic advantage. By focusing on specific, high-value tasks rather than trying to build general-purpose AI assistants, startups can:

1. Build Deeper Competitive Moats

• Domain-specific training data becomes a valuable asset
• Specialized optimization creates barriers to entry
• Closer alignment with customer needs

2. Deliver Better Results

• More reliable and consistent performance
• Faster response times
• Better error handling for edge cases

3. Create Sustainable Differentiation

• Clear value proposition
• Harder for competitors to replicate
• More defensible market position

The Path Forward: Right-Sizing AI Solutions

The next wave of successful AI startups won't be built on the premise of replacing humans with general-purpose AI. Instead, they'll focus on augmenting human capabilities with highly specialized, efficient AI tools. For startups looking to capitalize on this trend, the path forward requires a shift in thinking:

1. Start with the Problem, Not the Technology

2. Build for Sustainability

3. Embrace Specialization

By focusing on specialized, efficient solutions rather than trying to replicate general intelligence, startups can build more sustainable, effective, and valuable companies.