L7 Informatics CEO, Vasu Rangadass, Ph.D., examines the architectural principle separating pharmaceutical AI experimentation from enterprise-scale deployment.
AUSTIN, Texas — June 12, 2026 — Most pharmaceutical AI pilots never reach production, and according to L7 Informatics, the reason has almost nothing to do with the models themselves. It has to do with the operational environment that should have been built around them.
A recent architectural analysis of Anthropic's Claude Code found that only 1.6 percent of a production-grade AI system is actual decision logic. The remaining 98.4 percent is operational infrastructure: permission structures, context management, tool routing, audit, and recovery logic. That ratio, L7 Informatics argues, explains more about the current state of pharmaceutical AI than any conversation about model performance.
In pharmaceutical manufacturing, the requirements on these 98 percent are particularly demanding. The operational harness must enforce 21 CFR Part 11 and EU GMP Annex 11 compliance, maintain unbroken audit trails, manage the instrument connectivity and sample lineage that regulators expect, and govern the boundary between probabilistic AI reasoning and the deterministic execution that regulated environments require. An AI agent that cannot operate within those constraints is not a tool for pharmaceutical manufacturing, but a liability.
"Every large pharma organization I speak with is somewhere on the same arc," said Vasu Rangadass, Ph.D., President and CEO of L7 Informatics. "A promising AI pilot, real enthusiasm from the data science team, and then a slow stall when someone asks how it gets into production. The model works. The data works, more or less. But the surrounding environment does not, and no one quite planned for that. The organizations that will win pharma's agentic era are not the ones with the most sophisticated models. They are the ones that build the most capable operational harness."
Industry data reinforces the scale of the problem. McKinsey's 2025 State of AI survey found that while nearly nine in ten organizations now use AI regularly, 94 percent report no significant value from those investments. Gartner has projected that more than 40 percent of agentic AI projects will be canceled before the end of 2027, not because the technology fails, but because the governance and operational infrastructure surrounding it was never built.
L7 Informatics has structured its platform around the architectural principle that the intelligence should be thin and the harness should be robust. L7|ESP® provides the deterministic orchestration layer that enforces SOPs, manages GxP compliance gates, and generates the audit trail regulators expect. L7|SYNAPSE™ provides the agentic reasoning layer, integrating foundation models including Claude and Gemini as bounded reasoning engines whose outputs pass through deterministic enforcement before reaching the laboratory or manufacturing floor. The principle behind the architecture, what L7 describes as validate once, execute millions, concentrates the validation burden in a deterministic harness, making enterprise-scale AI economically viable in regulated environments.
Rangadass argues that the conversation pharmaceutical leaders should be having now is not about what AI can do, but about the accountability infrastructure required to let it do anything at all in a regulated setting. "Autonomy in any regulated system has to be earned through demonstrated accountability," he said. "Regulators, quality organizations, and ultimately patients need to be able to answer the question: when this system made a decision, how do we know it was the right one, and how do we trace what happened if it was not? That question cannot be answered by pointing to a model's performance on a benchmark. It can only be answered by pointing to a system that maintains a complete, auditable, validated record of every consequential action."
For the full argument, including the economic case for hybrid orchestration and a detailed view of how the operational harness applies across the pharmaceutical lifecycle, download the white paper Orchestrating the 98%: Why the Operational Harness Will Define Pharma's Agentic Era at https://l7informatics.com/white-paper/orchestrating-the-98-percent/
About
L7 Informatics L7
Informatics is on a mission to accelerate the pace of science. L7|ESP®, our
Enterprise Science Platform, is the operational backbone of modern life
sciences organizations, orchestrating people, instruments, data, and AI agents
that drive research, development, and manufacturing. Built for regulated
environments, L7|ESP unifies laboratory, manufacturing, and operational
workflows into a single execution layer, giving science the infrastructure it
needs to move faster and comply by design. L7|SYNAPSE™, L7|ESP's agentic AI
layer, brings reasoning and autonomous execution to that environment while
preserving the compliance and auditability that regulated operations require.
