The 'harness' built around an AI model is now as important as the model itself for creating powerful applications, marking a shift from a 'model is everything' perspective.
The demand for AI compute is highly elastic; significant cost reductions consistently lead to a more-than-proportional increase in usage, indicating a vast market limited primarily by price.
AI's application in fundamental scientific research, especially drug design and discovery, is an underappreciated domain with the potential for profound economic and technological progress.
A primary obstacle to enterprise AI adoption is accessing undocumented 'tribal knowledge' held by employees, which is not found in formal knowledge bases.
Developing specialized models, like GPT-5.1 for speed and Codex for coding, is a crucial strategy for addressing specific user pain points and driving adoption in targeted domains.
▶The Economics of AI ComputeApr 2026
This theme centers on the relationship between the cost of AI inference and market demand. Goodman asserts that demand is highly price-elastic, with every price cut by OpenAI resulting in a more-than-proportional increase in usage volume. He anticipates this trend will continue, with costs dropping by 'multiples' in the near future.
For investors, this suggests that the primary growth vector in the AI sector is not just model capability but also compute efficiency, and companies that can drive down inference costs will capture a disproportionate share of a rapidly expanding market.
▶Beyond the Model: The Rise of the 'Harness'Apr 2026
Goodman articulates an evolution in his thinking, moving from a 'model is everything' perspective to one where the 'harness'—the ecosystem of prompts, reference architectures, and surrounding engineering—is equally vital. He argues that simply swapping models via an API is no longer feasible for complex tasks and that providers will need to offer the harness alongside the model.
This signals a maturation of the AI market, where competitive advantage will shift from raw model performance to the usability and effectiveness of the entire application stack, creating opportunities for MLOps and AI implementation service companies.
▶AI as a Catalyst for Scientific DiscoveryApr 2026
Goodman repeatedly highlights the transformative potential of AI in scientific research, framing it as an 'underhyped' area with massive potential. He provides concrete examples, such as GPT-5 Pro replicating weeks of a physicist's work in minutes, and points to drug design and discovery as a key area for future impact.
Analysts should monitor AI adoption within R&D-heavy sectors like pharmaceuticals and materials science, as breakthroughs here could create entirely new markets and disrupt established players far beyond the tech industry itself.
▶Enterprise Adoption and the 'Last Mile' ProblemApr 2026
This theme explores the practical challenges and timelines for AI adoption in business. Goodman predicts 2025 will be the 'year of coding in the enterprise' but also identifies a key bottleneck: critical operational knowledge is often undocumented and resides only in employees' minds, making it difficult for AI systems to access.
This highlights a significant opportunity for solutions focused on knowledge capture and management as a prerequisite for successful enterprise AI deployment, suggesting that the market for AI-enablement tools may be as large as the market for AI models.