The current SaaS/AI narrative is wrong

No, this isn't a clickbait.

I trained Claude to have conversations with me in a particular manner. I simply reproduce my argument on this topic with Claude. It's fun this way that to read a monologue from me.

Claude said,

Given your experience as a repeat AI company founder and current focus on AI workflows, let me engage with your thesis directly.

I agree with most of your assessment but see one potential counterpoint worth exploring:

The moat isn't in the AI-powered workflow itself, but could emerge from three compounding elements:

1. Data network effects from real workflows: Each customer interaction generates unique examples of edge cases, errors, and industry-specific nuances. While individual tasks (like medical billing) may be "shallow," the accumulation of these edge cases across thousands of real customer interactions creates a dataset that's hard to replicate.

2. Task-chaining optimization: The value may lie not in replacing individual SaaS functions, but in understanding how these functions should optimally chain together in industry-specific contexts. For example, how a medical billing workflow integrates with insurance verification, patient communication, and regulatory compliance.

3. Industry-specific trust and compliance infrastructure: Particularly in regulated industries, the moat could be in building compliant infrastructure (audit trails, explanability, industry certifications) that's harder to replicate than the core AI functionality.

However, I agree that without these elements working together to create deep vertical specialization, the core AI workflow replacement will be commoditized. The "shallow" vertical context alone isn't enough.

Would you like to explore any of these potential moat elements in more detail? I'm particularly interested in your thoughts on whether the data network effects from real customer interactions could be substantial enough to matter.

Claude said,

As a strategic advisor and given your deep experience building multiple companies, I concur with your counterpoints - they effectively dismantle the traditional moat arguments.

Your point about data ownership is particularly incisive. The AI companies are essentially service providers to the real data owners (medical billing companies, law firms, etc.). This creates a fundamental weakness in any data network effect argument.

This leads to an interesting strategic question: If these businesses are valid but won't have lasting moats, what's the optimal strategy for a company like Moative? I see three potential approaches:

1. Speed to exit: Build and sell quickly before commoditization hits
2. Continuous innovation: Stay ahead by constantly expanding into new verticals/workflows, accepting that each individual solution will be commoditized
3. Platform play: Focus on becoming the infrastructure layer that powers multiple AI workflow companies, rather than building the workflows themselves

Claude said,

Your argument about interdisciplinary convergence is compelling because it solves several problems with the pure AI workflow replacement thesis:

1. Creates genuine complexity barriers:

- The coordination of multiple deep domains (like fisheries + AI + aquaculture) creates natural moats

- The knowledge integration challenge itself becomes a competitive advantage

- Competitors can't simply throw more AI at the problem

2. Addresses market reality:

- Regulated/high-stakes industries won't accept pure AI solutions

- Human expertise provides confidence and compliance coverage

- Creates natural buy-in from institutional stakeholders

However, potential holes in this thesis:

1. Scaling Challenge:

- Finding and coordinating domain experts across multiple fields is inherently difficult

- Each new vertical requires building new expert networks

- The human component makes it harder to maintain consistent quality at scale

2. Time Window Risk:

- As AI capabilities advance, will the need for human domain experts diminish?

- Could future AI models become sophisticated enough to integrate multiple domains without human experts?

- What's the time horizon for this advantage?

3. Economic Model Tension:

- Expert-driven models typically have higher costs and lower margins

- How do you price these solutions to reflect the value while maintaining competitiveness?

- Can this model attract venture capital that typically seeks higher margins?

......

IMO, AI puts deep tech on steroids, but the inter-disciplinary nature makes it a harder business to build than SaaS but that is the point. We need to solve harder problems as easier problems will be solved a million small AI craftists. Unlike SaaS, the supply of talent in these deep industrial expertise areas are locked in big companies, research labs, and academia. The industry will have to prove the viability of the path with some big successes before the academia reorients itself to produce more industry-stack software engineers (Marine Data Scientists?). We need the viability proofs to stop bioinformatics graduates from going to hedge funds. It will happen over a few years. A generation of big, audacious, industry upending companies will be built around this convergence, until that advantage fades away due to more supply of talent and better intelligence. But at least that won't happen next week if OpenAI releases O4.

PS: Moative puts these thoughts into action by working with exciting, industry leading companies that are very invested in understanding and owning their place in the new world order. Know a paranoid board? We have some camomile tea bags. Over tea, we may find some answers.