Slate: Moving Beyond ReAct and RLM
Source: https://randomlabs.ai/blog/slate Author: RandomLabs Date: 2026-03-13
Summary
RandomLabs introduces Slate — a new agent architecture that uses “threads” (persistent memory chains) instead of the standard ReAct loop. Argues ReAct is fundamentally limited because it has no persistent memory between runs; Slate addresses this.
Key Claims
- ReAct limitation: each ReAct cycle starts fresh — no memory of previous reasoning steps carries over to the next task.
- Thread concept: a thread is an active memory context that persists and evolves. Agents write to threads; other agents or the same agent reads them.
- Multi-agent benefit: threads allow multiple agents to share and coordinate via a common memory substrate.
- Slate vs. RLM (Recurrent Language Model): RLM tries to add memory via recurrence; Slate uses external state explicitly.
- Use case fit: tasks that require sustained focus over multiple sessions (week-long projects) benefit most from threads.
Concepts
- Agent Memory — threads as persistent agent memory
- Coding Agents — multi-agent coding coordination
- Context Engineering — threads as structured context management