To ensure product quality, Fixer pitted its AI against 10 of its own human executive assistants on the same tasks. They refused to launch features until the AI could consistently outperform the humans on accuracy, using their service business as a direct training and validation engine.
Related Insights
Effective enterprise AI deployment involves running human and AI workflows in parallel. When the AI fails, it generates a data point for fine-tuning. When the human fails, it becomes a training moment for the employee. This "tandem system" creates a continuous feedback loop for both the model and the workforce.
Instead of waiting for AI models to be perfect, design your application from the start to allow for human correction. This pragmatic approach acknowledges AI's inherent uncertainty and allows you to deliver value sooner by leveraging human oversight to handle edge cases.
Beyond automating 80% of customer inquiries with AI, Sea leverages these tools as trainers for its human agents. They created an AI "custom service trainer" to improve the performance and consistency of their human support team, creating a powerful symbiotic system rather than just replacing people.
The frontier of AI training is moving beyond humans ranking model outputs (RLHF). Now, high-skilled experts create detailed success criteria (like rubrics or unit tests), which an AI then uses to provide feedback to the main model at scale, a process called RLAIF.
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.
Unlike deterministic SaaS software that works consistently, AI is probabilistic and doesn't work perfectly out of the box. Achieving 'human-grade' performance (e.g., 99.9% reliability) requires continuous tuning and expert guidance, countering the hype that AI is an immediate, hands-off solution.
Do not blindly trust an LLM's evaluation scores. The biggest mistake is showing stakeholders metrics that don't match their perception of product quality. To build trust, first hand-label a sample of data with binary outcomes (good/bad), then compare the LLM judge's scores against these human labels to ensure agreement before deploying the eval.
To improve the quality and accuracy of an AI agent's output, spawn multiple sub-agents with competing or adversarial roles. For example, a code review agent finds bugs, while several "auditor" agents check for false positives, resulting in a more reliable final analysis.
Instead of generic benchmarks, Superhuman tests its AI models against specific problem "dimensions" like deep search and date comprehension. It uses "canonical queries," including extreme edge cases from its CEO, to ensure high quality on tasks that matter most to demanding users.
Instead of waiting for external reports, companies should develop their own AI model evaluations. By defining key tasks for specific roles and testing new models against them with standard prompts, businesses can create a relevant, internal benchmark.
