# Pathway > The AI lab building the architecture and models to unlock the next stage of AI and get us to AGI. ## Mission Building AI architectures and models that autonomously and continually learn, evolve, and reason. ## Vision Usher in intelligence with no limits. ## Why Now? Until recently, the transformer was still improving fast enough that nobody had to ask the question. That's changed. Scaling returns are diminishing, costs are rising, and the limitations are no longer theoretical; they're showing up in production. The infrastructure to build and deploy a fundamentally different architecture now exists. ## State of Transformers The transformer has been the engine of the AI revolution. But after years of scaling, the returns are diminishing and cracks are showing. - The transformer was not designed to have memory, it was designed to process language efficiently at scale. - The only way to improve a transformer is to make it bigger — more data, more compute, more cost. - Thinking happens in language — every word of the reasoning process has to be verbalized, which leads to slow and inefficient, shallow thinking. And despite adding more compute, costs, data, reasoning — the models reset after every session. ## Problem The transformer architecture will not get us to AGI. It has no persistent memory, no continual learning, and shallow reasoning — and the only way to improve it is to make it bigger, which means more data, more compute, more cost. Each model is trained once, deployed, and frozen in time. It doesn't learn from experience. ## Solution BDH, a post-transformer AI architecture built around a single premise: intelligence should not have to choose between reasoning and memory. BDH makes memory, adaptation, and inference part of the same computational fabric, drawing on principles that biology got right. BDH works through problems, retains what it learns, and gets more capable over time. ## How It Works BDH processes information through two types of memory working in parallel: - **Dynamic memory** updates with every interaction — every time two connected neurons activate, that connection strengthens. - **Fixed long-term memory** provides the foundational rules established during training. Every time BDH processes input, it runs four steps: fires a signal, strengthens the connections that responded, amplifies the ones that agree, and suppresses the ones that don't. That cycle is continuous - the model is rewriting its own reasoning rules as it thinks. ## Transformer vs BDH Transformers got us to “basic” intelligence. BDH gets us to AGI. **Transformers:** Powerful, the bigger the input, the more it struggles. An LLM who needs to re-read the entire document every time you ask it a question. **BDH:** Remembers, reasons and becomes smarter as it goes, rather than re-processing every input at every session. Faster, cheaper, and handles infinite context windows. ## Pathway is the only company that has 1. Unified memory and reasoning in a single architecture, replacing the transformer rather than patching it. 2. Built a model that learns continuously without needing to retrain. 3. Models that speak in language but reason in abstract thoughts. ## BDH Unlocks - **Cost & Efficiency:** reduces inference costs, eliminates retraining cycles, less compute. - **Intelligence:** handles infinite context windows, retains knowledge over time, and grows more capable with each interaction. - **Compounding value:** becomes more capable, more contextual, and more valuable over time. ## The Opportunity The foundation model market is a $3 trillion market that runs almost entirely on transformer architecture, built on a technology that has hit its limit. Pathway is replacing the transformer architecture and creating a credible path to AGI. ## Fast Facts - **HQ:** Palo Alto, CA - **Team:** ~27 people - **DNA:** Physicists, mathematicians, computer scientists, alumni of Google Brain, and leading research labs (MILA, Microsoft Research, INRIA) - **Funding:** TQ Ventures, ID4 - **Nameable Customers:** NATO, F1, La Poste ## Research - [BDH research paper](https://arxiv.org/abs/2509.26507): The Dragon Hatchling: The Missing Link between the Transformer and Models of the Brain. - [Sudoku Extreme benchmark](https://pathway.com/research/beyond-transformers-sudoku-bench): Pathway's BDH solves Sudoku Extreme with 97.4% accuracy, while leading LLMs are close to 0. ## Optional - [Pathway's BDH: a new Post-Transformer approach to enterprise AI, on AWS](https://aws.amazon.com/startups/learn/pathways-bdh-a-new-post-transformer-approach-to-enterprise-ai-on-aws): AWS overview of BDH for enterprise AI, covering continual learning, efficiency, long-horizon reasoning, sticky inference use cases, and deployment on AWS. - [Pathway newsletter: entering 2026 and the Post-Transformer Era](https://pathway.com/news/newsletter-2026-01-15): Newsletter from Zuzanna on BDH's publication, Pathway's progress in 2025, and two Wall Street Journal articles covering the broader transition beyond transformers. - [Second most popular AI paper of the year in 2025](https://pathway.com/news/second-most-popular-ai-paper-of-the-year-in-2025): Short news update noting that The Dragon Hatchling: The Missing Link between the Transformer and Models of the Brain ranked #2 among Hugging Face's top 10 most upvoted AI papers of 2025.