SAN FRANCISCO, CA – May 05, 2026 – In a bold move against leading AI giants, the San Francisco-based startup Causal Dynamics Lab (CDL) has launched an innovative product that surpasses the coding models from OpenAI and Anthropic across multiple key metrics. Their new offering, Cielara Code, addresses a significant and costly flaw in today’s AI coding assistants: their inability to grasp the context of the software they modify.
Though AI coding tools are capable of generating code at remarkable speeds, this efficiency comes at a cost. The 2025 DORA report, an authoritative industry measure of software performance, indicated a 7.2% decline in deployment stability directly linked to the use of AI coding tools. This escalating concern, termed “dynamic verification debt” by AWS CTO Werner Vogels, underscores a critical divide between an AI’s ability to write code and its capacity to comprehend the implications of that code in a live production setting.
Rather than aiming to replace models like OpenAI’s Codex or Anthropic’s Claude Code, CDL’s Cielara Code is designed to serve as an essential safety layer that enhances these existing technologies. Preliminary benchmark results indicate that this approach is not only effective but also superior in addressing one of the most challenging aspects of software development.
The Limitations of AI Understanding
Research conducted by Causal Dynamics Lab identifies the main restraint for current AI coding agents as a significant lack of awareness. After analyzing thousands of AI-driven coding sessions, CDL discovered that a startling 56.8% of an agent’s time was spent merely reading files, while another 24.2% involved using basic search commands like grep. Less than 1% of their actions were dedicated to the primary task: editing code.
The findings indicate that AI agents are often navigating in the dark, treating complex software architectures as simple text. They struggle to understand file interconnections, function calls, or the ripple effects of changes made in one area. This issue worsens in large, intricate codebases; CDL’s research found that when a fix required adjustments to more than six files, the agents’ ability to recall necessary information dramatically decreased, while the compute resources wasted on unsuccessful attempts surged fourfold.
“Every coding agent currently uses grep, akin to a surgeon operating without imaging,” remarked Hasibul Haque, CEO of Causal Dynamics Lab and former head of platform engineering at Uber. “With Cielara Code, we aimed to enhance the AI’s vision of its environment, making the rationale behind every change transparent and verifiable.”
This concern is not merely theoretical. A well-documented issue on GitHub (#42796) for Claude Code exemplifies how current agents struggle to grasp the interconnected nature of modern software, often resulting in flawed or incomplete solutions.
Equipping AI with a Comprehensive Map
Cielara Code’s innovation lies in its unique strategy for representing software. Instead of allowing an AI agent to meander through a directory of files, it first constructs a detailed map of the entire system through a proprietary “Production World Model.” This model is visualized as a six-layer causal graph.
This graph goes beyond a standard dependency tree. It encapsulates profound contextual details about the software, encompassing what the code does, its purpose, its ownership, operational restrictions, execution environments, and runtime behaviors. When a failure occurs, the system can trace the issue back not only to the specific line of code that caused the problem but also to the developer who approved it and the business rationale for the change.
This contextual mapping enables Cielara Code to guide AI agents with unmatched accuracy. In three independent benchmarks, it achieved an overall code localization accuracy of 0.774, outpacing Claude Code (Opus-4.6) at 0.738 and OpenAI Codex (GPT-5.4) at 0.707. Notably, in the MULocBench test, which encompasses over 1,000 issues, Cielara reduced task time while cutting computing costs by 30 to 40 percent.
The technology driving this innovation is REASONARA, a graph-structured causal memory layer capable of storing an extensive context of over 125 million tokens. Yet, it intelligently retrieves only the handful of tokens relevant to a specific task, achieving a context-lookup reduction of up to 98% compared to conventional methods. This efficiency lends itself to quicker and more effective analyses.
Enterprise Adoption and Risk Management Initiatives
The market’s response has been swift. Cielara Code is already being employed by 11 Fortune 100 companies and over 40 Fortune 500 firms, who regard it as a vital component of their increasingly automated development infrastructures.
For enterprise leaders, this tool alleviates mounting concerns. “Board members and auditors expect more proactive risk management,” said the Chief Information Security Officer of one of the largest law firms in the United States, a customer of Cielara Code. “Leaders want assurance that security measures can anticipate risks posed by rapid AI and automation, rather than merely reacting post-incident.”
This sentiment resonates among other industry figures who view the technology as an essential evolution. Phillip Miller, Vice President and Global CISO at H&R Block, described CDL’s inventions as a “generational leap toward the original promise of AI.” He added, “Enterprises require solutions for challenges that human effort alone cannot address… My book, Hacking Success, outlined a future where AI needs clear, directive policy—not just rules or guardrails—to operate safely and effectively. Now, businesses have the opportunity to harness Cielara’s models for overseeing AI deployments and their supporting infrastructure.”
A New Era for AI Development
Causal Dynamics Lab, established by a team comprising veterans from Uber’s platform and AI researchers from Microsoft Research and Emory University, views its current offerings as just the beginning. The foundational Production World Model is designed to serve a much larger vision.
“AI has already revolutionized how people access information,” noted Matt Fisher, former Co-Founder and CTO of Daydream. “The next phase is transforming how people make decisions by exploring alternatives, weighing options, and understanding outcomes before taking action. That shift toward exploring consequences is where CDL is aiming to head.”
The company’s roadmap includes expanding its simulation capabilities to foresee the entire impact of changes encompassing not just code, but also infrastructure, policy, and operations. Ultimately, their ambition is to create a permanent, enterprise-wide reasoning layer that any AI agent can consult prior to implementing changes, ensuring that the rapid pace of AI development is balanced by appropriate safety and comprehension.
