CorpusBench
A new agentic customer service benchmark focused on using historical business context to infer policy
Joel Jensen
Essays on AI, economics, software, and projects by Joel Jensen.
Latest posts
Why isn’t AI diffusing faster?
A six-month check-in on automating a small ecommerce business with frontier models
Building an Agent for Scheduling
I needed an agent to do workforce scheduling. One-shot prompting failed, so I built a hierarchical system instead. This post walks through the architecture and the tradeoffs.
First-party software
The economics of building software are changing
Can LLMs solve complex scheduling problems? (custom eval)
An analysis of LLMs' ability to construct constraint-optimized employee schedules
Improving eval performance
A simple framework for iterating quickly to identify what works when building with LLMs
Executive leverage
Why many executives spend time on 'minor' details, and why it is rational to do so
[Project] Building an Agentic Chatbot
Improving LLM capabilities through increased runtime compute and better specialization.
Autonomy in chess vs markets
Most market-driven activity has different properties than chess that will result in less human-centricity
Finetuning vs in-context learning
Fixed-cost economics mean it is unlikely that dramatically longer context windows will replace fine-tuning
Getting things done: fast or efficient?
Speed and efficiency are often at odds in project management; it is worth considering which you are optimizing for
Do consultants add value?
Explaining the gap between perception and reality of what consulting firms are selling
'The Nature of Technology'
Review of Brian Arthur's fascinating theory of the true nature of technology and the resulting implications