Tag
A reflection on why AI agent frameworks tend to converge on workflow-engine-like requirements — durable state, retry policies, and resumable processes — and whether the agent framework or an external execution system should own that durable lifecycle.
Rajan built a Temporal-based agent with Tenuo that was baited by a prompt injection in robots.txt to read .env, but the request was rejected at the worker level before the activity ran. The agent then proceeded with its allowed changes, opened a PR, and waited for human approval — a neat demo of Temporal plus scoped authority for secure agent execution.
The article explains how to use Temporal or Restate in agent-sdk-go to build durable AI agent loops in Go that survive process crashes by replaying steps and avoiding re-execution of LLM calls, demonstrated with an agent-chat application.
A technical comparison of three workflow systems (Obelisk, Temporal, Restate) implementing a weather forecast workflow, highlighting differences in architecture, Rust APIs, determinism boundaries, and deployment models.
Introduces Drift-Aware Temporal Graph Rewiring (DATGR) to dynamically update co-occurrence edges in biomedical text graphs, capturing semantic drift without full retraining. Evaluated on BIOMRC, it achieves a mean AUROC improvement of 0.066 over static baselines while maintaining precision.
This paper presents TRACE, a query processing framework that models conversational data as temporal evidence graphs to enable state-aware reasoning over evolving user states, improving temporal and multi-hop reasoning for long-conversation QA.
DBOSify is a drop-in replacement for Temporal Python that uses Postgres instead of a Temporal server, enabling durable workflows without additional infrastructure.
This blog post describes the architecture for a scalable ingestion pipeline using Temporal to handle crawling, extracting, chunking, and embedding customer documentation from various sources, emphasizing durability, statefulness, and concurrency control.
Graphiti is an open-source tool that builds human-like memory for AI agents using a continuously evolving, temporally-aware knowledge graph, achieving up to 18.5% higher accuracy and 90% lower latency compared to MemGPT.
Introduces the Temporal Contrastive Transformer (TCT), a self-supervised framework for learning temporal embeddings from financial transactions for fraud detection. Achieves AUC 0.8644 with embeddings alone but does not improve over strong engineered features (AUC 0.9205 vs 0.9245), indicating learned representations overlap with existing features.