Baseten discussion explores how to make agents reliable over hours and days
baseten · x · 2026-07-23
A long discussion with Baseten’s Gabe Pereyra focuses on what it takes to build agents that can reliably complete work over hours, days, or longer. The conversation highlights why today’s agents struggle with search and long context, and points to techniques such as KV-cache compaction, synthetic data, and continual learning.
- Main question: how to make long-running agents reliable.
- Key bottlenecks: search and context-length limits.
- Proposed directions: KV-cache compaction, synthetic data, continual learning.
Related event: Harvey and Baseten Discuss Technical Challenges of Long-Running Agents(3 posts)→
More from coding & agent
- Gemini 3.5 Flash-Lite is 71x Cheaper Than Claude for Doc Extraction — rseroter · 2026-07-23
- Cohere to Host Talk on LLM Agent Reliability & Uncertainty Signals — Cohere_Labs · 2026-07-23
- LangChain and Cognition will host a meetup on open memory for agents — LangChain · 2026-07-23
- Factory Co-founder Predicts 90% of Coding Agent Tokens Will Be Fully Autonomous in 12-24 Months — matanSF · 2026-07-23
- Voice assistant tool calls sped up instantly after moving the backend to Europe — ur_piyo_a_hoe · 2026-07-23
- W&B’s Scott Condron wants to push research agents, trace insights, and marimo eval UIs — _ScottCondron · 2026-07-23