· 4 min read
The More Your AI Remembers, the Harder Memory Becomes
As AI agents accumulate years of data, they risk becoming rigid or inaccurate. Effective memory requires context, expiration, and the ability to adapt to change.
Exciton Research
Blog
Essays and research notes on AI agents, evaluation, and the engineering decisions behind reliable systems.
· 3 min read
Automatic AI Agent Evaluation: What to Check with Code, Reference Data, and LLM Judges
Learn how to evaluate AI agents by combining automated code checks, reference data, and LLM judges to measure task success, accuracy, and safety effectively.
· 1 min read
A Checklist for Evaluating AI Agents
Evaluate AI agent performance with a structured checklist covering task success, safety, tool use, and error recovery to ensure reliable, secure deployment.
· 2 min read
Beyond Git: How Anthropic Is Redesigning Source Control for AI Agents
Anthropic's Scion system redesigns source control for AI agents, moving from distributed Git to a server-centric architecture to handle massive coding workloads.
· 3 min read
How to Evaluate AI Agents
Measuring agent performance requires verifying complex, multi-step actions. Learn how to design evaluations that capture reliability, cost, and true success.
· 4 min read
Securing Agents That Act for You
When agents browse, remember, and transact, they face unique security risks. You must isolate external data from internal authority to prevent misuse.
· 2 min read
Managing AI-Discovered Vulnerabilities: Lessons from the PHP2Shell Incident
Recent AI-driven discovery of critical PHP vulnerabilities highlights a shift in security. Learn how to adapt your patching processes and agentic workflows.