Our Research

Research

Insights and findings from our work on AI agent optimization, evaluation methodology, and the science of building reliable autonomous systems.

July 27, 2026

New!

A Data-Driven Explanation: Why Do AI Agents Still Fail

AI agents were projected to create $450 billion in value by 2028, yet enterprise deployment is stuck in the single digits. A data-driven look at the four things behind the gap: variance, benchmarks, agent errors, and alignment.

Read article

January 29, 2026

Loss of First Principles Science with Agentic AI

There is an innate human desire for immediate causality. Software engineering is built on that feeling: write a test, find the bug, fix it, watch it pass. In the stochastic world of AI agents, that loop stops working the way it used to...

Read article

January 14, 2026

Making GPT-4.1-mini as Good as GPT-5.2

We improved GPT-4.1-mini from 53% to 72% on τ²-bench using Lucidic's optimization algorithms, achieving performance comparable to GPT-5.2 at a fraction of the cost.

Read article