Self-Optimizing Agents and the Fast-Moving Enterprise Stack

Events

Report: Salesforce Agentforce hit $1.2B ARR, accelerating faster than initial cloud launches

This matters for builders because Agentforce isn't just a tool; it's becoming Salesforce's primary revenue driver. The platform is where enterprises are buying agents, so this is where the money is flowing first.

Research

API Kittens and the limits of LLM tool-calling: How API documentation mistakes cripple agents

This paper finds that the architecture of your tool definitions dramatically impacts agent capability. For builders, it means API docs aren't just reference material; they are the driver of agent performance.

Self-optimizing agent architecture enables real-time code modification

This work explores letting agents rewrite their own logic mid-task. The implications for developer oversight are massive, but so is the potential for building agents that actually teach themselves to plug their own gaps.

Cognitive architecture for LLM agents: A deep dive into memory and retrieval

The focus here is memory and retrieval: making sure your agent isn't effectively starting from scratch every time it turns over a task. We need to build systems that actually trust past context.

Tools

Firecrawl launches "Turn websites into LLM-ready data" with agent-focused architecture

Firecrawl shipped privacy-respecting browser agents that act as the eyes for your own AI agent stack. It signals the move toward vertically integrated infrastructure where the crawler is built specifically to plug into an LLM.