Subvolts take New AI memory tech could solve LLM context limits, enabling better agents and RAG. Nov 9, 2025 - Developers
- Learn about new memory architectures that can improve context retention and enable more reliable AI agents and RAG systems.
- Business
- Understand how advancements in AI memory can lead to more capable agents, better customer service, and improved data analysis tools.
This interview explores a promising new architecture for AI memory that could significantly improve LLM capabilities for long context tasks.
#299 Jacob Buckman: Why the Future of AI Won't Be Built on Transformers
Audio streams from Craig S. Smith. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
Craig Smith hosts Eye on AI episode #299 with Jacob Buckman of Manifest AI. They discuss Manifest AI's Power Retention architecture, which aims to solve LLM memory limitations. The conversation explores how new models can retain context across long conversations and documents, moving beyond transformer limitations and offering practical paths for building more capable AI agents and RAG systems.
Imagine AI that remembers everything you say, even in very long chats. This episode explains new ways to build AI that doesn't forget, making it much more useful for tasks like understanding long documents or having extended conversations.
This episode is for AI developers and product leaders building agents and RAG systems. It explains how new memory architectures can overcome transformer limitations for long context. Listeners will learn about practical implementation, metrics for evaluating long context quality, and how to design for reliable recall and task completion. This research directly addresses current challenges faced by frontier labs like Google Gemini and Anthropic in scaling context windows effectively.
- HostCraig S. Smith
- GuestJacob BuckmanManifest AI
- Episode Title: Why the Future of AI Won't Be Built on Transformers
- Show: Eye on AI
- Guest: Jacob Buckman
- Guest Organization: Manifest AI
- Key Concept: Power Retention architecture for AI memory
- Metrics Discussed: recall consistency, answer fidelity, task completion, CSAT, cost per resolution
This episode is sponsored by AGNTCY. Unlock agents at scale with an open Internet of Agents.
Visit https://agntcy.org/ and add your support.
Why do today's LLMs forget key details over long context, and what would it take to give them real memory that scales?
In this episode of Eye on AI, host Craig Smith explores Manifest AI's Power Retention architecture and how it rethinks memory, context, and learning for modern models. We look at why transformers struggle with long inputs, how state space and retention models keep context at linear cost, and how scaling state size unlocks reliable recall across lengthy conversations, code, and documents. We also cover practical paths to retrofit existing transformer models, how in context learning can replace frequent fine tuning, and what this means for teams building agents and RAG systems.
Learn how product leaders and researchers measure true long context quality, which pitfalls to avoid when extending context windows, and which metrics matter most for success, including recall consistency, answer fidelity, task completion, CSAT, and cost per resolution. You will also hear how to design per user memory, set governance that prevents regressions, evaluate LLM as judge with human review, and plan a secure rollout that improves retrieval, multi step workflows, and agent reliability across chat, email, and voice.
Stay Updated:
Craig Smith on X:https://x.com/craigss
Eye on A.I. on X: https://x.com/EyeOn_AI
- LLMs
- Jacob Buckman
- Large Language Models, the AI systems that power chatbots and text generation.
- Transformers
- Jacob Buckman
- A common type of AI model architecture that is good at processing sequential data like text, but struggles with very long inputs.
- RAG systems
- Jacob Buckman
- Retrieval Augmented Generation systems, which combine AI models with external data to provide more accurate answers.
55 min42 min51 min34 minUkraine's Secret Weapon: The Points System Winning the Drone War | Andrii Hrytseniuk
WithAndrii Hrytseniuk
Same guests, other shows
57 min
Questions people ask
- Why do current LLMs struggle with long context?
- Current LLMs, often built on transformers, struggle to retain key details over long inputs. This leads to forgetting important information in extended conversations or documents.
- What is Manifest AI's Power Retention architecture?
- It is a new approach to AI memory that rethinks how models handle context and learning, aiming to provide reliable recall across lengthy interactions and data.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Craig S. Smith.