Beyond Stateless Prompts: Building Persistent, Context-Aware AI Agents
Abstract
Most AI agents are stateless. Switch context or close the window and the agent forgets everything, which returns the person to the beginning and puts the cost of continuity back onto them. This paper sets out architectural patterns for building persistent, context-aware agents that maintain memory across task switches and update their capabilities dynamically without being re-prompted. The distinction it draws is between an agent that is given context each time and one that holds it, since only the second can be trusted with work that spans more than a single exchange. The patterns are demonstrated through an agent maintaining context across a real multi-tool workflow rather than in a single-turn example, which is where stateless designs usually appear adequate. Readers take away an implementable blueprint for moving an agent from a stateless, prompt-dependent tool to a persistent, production-ready assistant that retains what it is doing, described concretely enough to apply immediately.
- Published in
- Proceedings of the Raptors Conference, Vol. 1 (2026)
- Conference
- Raptors Conference 2026, 21 October 2026, online
- Accepted
- 2026-09-16
- DOI
- https://doi.org/10.68495/rc.2026.C047
- Presentation
- Slides are not yet available for this paper.
- Licence
- © 2026Naga Putta. Published under Creative Commons Attribution 4.0 International (CC BY 4.0).
- Publisher
- Hackathon Raptors, Community Interest Company 15557917, London, United Kingdom
- Permanent address
- https://proceedings.raptors.dev/putta-beyond-stateless-prompts-building
Open access. Every entry in the Proceedings is free to read, with no charge to authors and no charge to readers. Slides are served from this site, so they stay available at the paper's permanent address.
Cite this paper
Naga Putta (2026). Beyond Stateless Prompts: Building Persistent, Context-Aware AI Agents. Proceedings of the Raptors Conference, Vol. 1 (2026). Hackathon Raptors, London. https://doi.org/10.68495/rc.2026.C047
@inproceedings{raptors2026-c047,
title = {{Beyond Stateless Prompts: Building Persistent, Context-Aware AI Agents}},
author = {Putta, Naga},
booktitle = {Proceedings of the Raptors Conference},
volume = {1},
year = {2026},
publisher = {Hackathon Raptors},
address = {London, United Kingdom},
doi = {10.68495/rc.2026.C047},
url = {https://doi.org/10.68495/rc.2026.C047},
abstract = {Most AI agents are stateless. Switch context or close the window and the agent forgets everything, which returns the person to the beginning and puts the cost of continuity back onto them. This paper sets out architectural patterns for building persistent, context-aware agents that maintain memory across task switches and update their capabilities dynamically without being re-prompted. The distinction it draws is between an agent that is given context each time and one that holds it, since only the second can be trusted with work that spans more than a single exchange. The patterns are demonstrated through an agent maintaining context across a real multi-tool workflow rather than in a single-turn example, which is where stateless designs usually appear adequate. Readers take away an implementable blueprint for moving an agent from a stateless, prompt-dependent tool to a persistent, production-ready assistant that retains what it is doing, described concretely enough to apply immediately.}
}
TY - CONF TI - Beyond Stateless Prompts: Building Persistent, Context-Aware AI Agents AU - Putta, Naga T2 - Proceedings of the Raptors Conference VL - 1 PY - 2026 DA - 2026/09/16 PB - Hackathon Raptors CY - London, United Kingdom DO - 10.68495/rc.2026.C047 UR - https://doi.org/10.68495/rc.2026.C047 AB - Most AI agents are stateless. Switch context or close the window and the agent forgets everything, which returns the person to the beginning and puts the cost of continuity back onto them. This paper sets out architectural patterns for building persistent, context-aware agents that maintain memory across task switches and update their capabilities dynamically without being re-prompted. The distinction it draws is between an agent that is given context each time and one that holds it, since only the second can be trusted with work that spans more than a single exchange. The patterns are demonstrated through an agent maintaining context across a real multi-tool workflow rather than in a single-turn example, which is where stateless designs usually appear adequate. Readers take away an implementable blueprint for moving an agent from a stateless, prompt-dependent tool to a persistent, production-ready assistant that retains what it is doing, described concretely enough to apply immediately. LA - en ER -