With AI making code generation nearly free, the new competitive moat isn't speed of execution but the quality of judgment, architecture, and intent behind the product. Mindless, 10-minute builds are easily replicated and lack defensibility, leading to what Aishwarya Srinivasan calls "AI slop."
Evaluating agentic AI requires an end-to-end "whole system eval" that assesses every connection, tool, and potential failure point. This is crucial because the non-deterministic nature of these systems creates a compounding effect of uncertainty that can't be captured by only checking the final LLM output.
Reinforcement Learning (RL) is ideal for fine-tuning AI agents because it allows them to self-learn and align their behavior by exploring vast, non-deterministic environments. This is a more scalable approach than supervised fine-tuning, which would require an impossibly large, pre-curated dataset to cover all potential pathways.
The key AI skill is evolving from crafting individual prompts to "loop engineering." This means defining goals and feedback systems that enable an agent to generate, self-review, and autonomously refine its output to meet a specific objective, minimizing the need for constant human-in-the-loop intervention.
In the fast-moving AI space, the most critical career skill is the ability to rapidly experiment and update your own hard skills. This involves constantly testing new models on personal use cases to develop an intuitive "vibe evaluation" sense and quickly separating signal from noise. This applies to technical leaders as well as individual contributors.
AI tools provide an easy entry point to many skills, which becomes a "crutch" if users don't move beyond the basics. The real power comes from using AI as a "catalyst" to infinitely raise the ceiling of your capabilities, pushing past the low barrier to entry to achieve more than was previously possible.
Startups succeed in AI adoption through sheer speed, launching products quickly and openly asking users to find flaws. In contrast, large enterprises are hampered by slow governance and red tape, causing their AI products to be outdated by the time they navigate internal approvals and finally launch.
While early AI companies built moats on user data by being first-to-market with API wrappers, today's startups cannot. A defensible AI product now requires a true moat beyond a simple wrapper, as model providers can easily replicate basic functionality, making such businesses non-defensible.
Top founders balance strong vision with flexibility. The mantra "strong opinions, loosely held" is key. While passionate, they aren't so married to their initial idea that ego prevents them from pivoting when faced with evidence of failure. This agility is crucial for survival, especially in early stages.