AI struggles with long-horizon tasks not just due to technical limits, but because we lack good ways to measure performance. Once effective evaluations (evals) for these capabilities exist, researchers can rapidly optimize models against them, accelerating progress significantly.

As benchmarks become standard, AI labs optimize models to excel at them, leading to score inflation without necessarily improving generalized intelligence. The solution isn't a single perfect test, but continuously creating new evals that measure capabilities relevant to real-world user needs.

The primary bottleneck in improving AI is no longer data or compute, but the creation of 'evals'—tests that measure a model's capabilities. These evals act as product requirement documents (PRDs) for researchers, defining what success looks like and guiding the training process.

Building a functional AI agent is just the starting point. The real work lies in developing a set of evaluations ("evals") to test if the agent consistently behaves as expected. Without quantifying failures and successes against a standard, you're just guessing, not iteratively improving the agent's performance.

If all your evals pass, you don't know the current limits of your system. Evals that consistently fail act as a benchmark. When a new foundation model is released, rerunning these tests immediately reveals if it has overcome previous limitations.

Traditional AI benchmarks are seen as increasingly incremental and less interesting. The new frontier for evaluating a model's true capability lies in applied, complex tasks that mimic real-world interaction, such as building in Minecraft (MC Bench) or managing a simulated business (VendingBench), which are more revealing of raw intelligence.

The most valuable evals aren't built with complex software but are often simple spreadsheets. Their power comes from deep subject matter expertise, which is necessary to create nuanced prompts and accurate scoring criteria that truly test a model's ability in a specific domain like clinical genomics or law.

Many people struggle to define what 'good' looks like. Building an evaluation (eval) for an AI system requires you to codify your quality standards, forcing a level of clarity and commitment that improves your own process and the AI's output.

Evals are designed to guide improvement. A set that scores 100% is too easy and provides no optimization path. A 0% score is too hard and offers no signal. A 25-50% success rate creates a "Goldilocks" zone—a challenging but achievable target for engineering teams to iterate against.