Descriptions:
Doug O’Laughlin, founder of SemiAnalysis and one of the semiconductor industry’s most-followed independent research analysts, joins Latent Space for a conversation spanning two tightly linked themes: how Claude Code is reshaping professional information-work workflows, and what SemiAnalysis calls the global memory shortage — the structural HBM supply constraint shaping AI infrastructure buildout.
On the AI tools side, O’Laughlin offers an unusually specific practitioner account of deploying Claude and Claude Code for financial and technical analysis. He frames the current generation of agentic AI as a “junior analyst” — excellent at painful information-gathering and synthesis tasks, but lacking the meta-level pattern recognition that makes senior analysts valuable. Specific behavioral observations include context window degradation over long tasks (“it gets dumber over time”), consistent error rates requiring expert review at output, and a gap between task execution and genuine domain reasoning. Despite these limitations, O’Laughlin argues the productivity gains are already irreversible for information workers and predicts Claude Code will become a baseline skill in financial analysis within 24 months.
The memory shortage discussion draws on SemiAnalysis’s proprietary supply-chain research, covering HBM capacity constraints from SK Hynix, Samsung, and Micron, implications for AI training cluster scaling, and how memory bandwidth — not raw compute — is increasingly the binding constraint for large-scale inference. The combination of cutting-edge AI adoption and deep semiconductor expertise makes this a distinctive data point on how domain experts are actually integrating these tools.
📺 Source: Latent Space · Published February 24, 2026
🏷️ Format: Interview
