Jerry Liu argues evals can replace hand-built agent workflows for most tasks
Overview
Jerry Liu, in an X post, argues that most tasks can now be solved by defining an eval and hillclimbing on it instead of hand-coding a deterministic or agentic workflow.
He says developers should focus on defining goals and success measures while frontier models do the work, with data providers building evals across economic activity. Liu expects agent interfaces to reduce most tasks to goals and eval instructions, but says the most complex processes will still need explicit workflow builders. This is his prediction and opinion, not a measured result. Harrison Chase later posted a piece titled "eval driven development" that frames evals as a development approach, but the post itself offers no further detail, so it adds no independent evidence for Liu's claim.
Written by AI from the articles below · updated Oct 9, 4:44 AM ET
Check the sources:
Developments
2 developments
- Oct 9, 3:50 AM ET · 1 articleHarrison Chase on eval-driven development for AI agentsHarrison Chase: Harrison Chase on eval-driven development for AI agents
- Oct 9, 1:58 AM ET · 1 articleJerry Liu says evals now replace hand-built agent workflowsJerry Liu: Jerry Liu says evals now replace hand-built agent workflows
Article timeline
Follow the coverage from different perspectives. Times are ET.
- Harrison ChaseHarrison Chase on eval-driven development for AI agents
AIHarrison Chase's post is titled "eval driven development," presenting evals as a development approach. The main post gives no further detail beyond the title. The quoted context from Jerry Liu argues that most tasks can be solved by defining an eval and hillclimbing over it rather than hand-building a deterministic or agentic workflow.
- Jerry LiuJerry Liu says evals now replace hand-built agent workflows
AIJerry Liu argues that most tasks can now be solved by defining an eval and hillclimbing on it, rather than hand-coding a deterministic or agentic workflow. He says data provider companies are building evals across economic activity so frontier models can handle more work, leaving developers to define goals and success measures. He expects agent interfaces to compress most tasks into goals and eval instructions, while the most complex processes will still need explicit workflow builders.
Heat trend
Not enough continuous observations to show a trend yet.