Poetic

Poetic is an artificial intelligence research company developing systems that enhance the reasoning capabilities of large language models (LLMs). Founded in 2025 by former Google DeepMind researchers Shumeet Baluja and Ian Fischer, the company focuses on building an AI “meta-system” that operates on top of existing foundation models rather than creating new models from scratch. Its approach is centered on improving problem-solving performance, reasoning accuracy, and efficiency across a wide range of applications.

The company’s technology is based on recursive self-improvement, a framework in which AI-generated agents continuously refine their own reasoning strategies. Instead of relying on large-scale retraining or reinforcement learning pipelines that require vast amounts of data and computing resources, Poetic aims to create specialized reasoning systems using relatively small datasets. This allows organizations to develop task-specific AI agents more quickly and at lower cost.

Poetic’s platform is designed to work with leading foundation models from providers such as OpenAI, Anthropic, Google, and Meta. By adding an intelligence layer above these models, the company seeks to improve their ability to solve complex problems, synthesize information, and adapt to new domains without modifying the underlying models themselves.

The company emerged from stealth in 2026 after reporting strong results on advanced reasoning benchmarks designed to measure general problem-solving ability. Around the same time, it raised $45.8 million in seed funding to support product development and expansion.

Poetic is part of a growing category of AI infrastructure companies focused on improving the performance of existing foundation models through orchestration, reasoning systems, and agent-based architectures rather than pursuing larger standalone models.

Keywords: Poetic, artificial intelligence, AI reasoning, recursive self-improvement, AI agents, large language models, foundation models, AI infrastructure, machine reasoning, DeepMind alumni, agentic AI, AI optimization, enterprise AI, reasoning systems.