Agents need control flow, not more prompts

kept by eddie may 09

Building deterministic control flow into agent harnesses, rather than relying on prompts alone, is essential for reliable AI agent behavior.

Structured Outputs Create False Confidence
Structured outputs via constrained decoding force LLMs to prioritize format conformance over response quality, degrading extraction accuracy and suppressing useful error signaling.
also on BAML, #software-engineering, #developer-tools
The Harness Is the Company - by Shrivu Shankar
SaaS businesses will evolve into AI-driven 'harnesses'—systems of infrastructure, context, and agents—that invert human roles from operators to taste-holders within the organization.
also on LangGraph, #ai-agents, #ai-coding, #software-engineering, #developer-tools
Open SWE: An Open-Source Framework for Internal Coding Agents
LangChain releases Open SWE, an open-source framework for building internal coding agents that mirrors patterns from Stripe, Ramp, and Coinbase production deployments.
also on LangGraph, #ai-agents, #ai-coding, #software-engineering, #developer-tools
The Death of Software 2.0 (A Better Analogy!)
AI agents like Claude Code will make human-oriented software obsolete, shifting value to persistent data infrastructure and APIs, analogous to a memory hierarchy.
also on Salesforce, #ai-agents, #ai-coding, #software-engineering, #developer-tools
Extensible Software in the age of LLMs
Extensible web software can combine a stable core, LLM-driven extensions, and capability-based sandboxes to let users safely create and share personalized features.
also on Salesforce, #ai-agents, #ai-coding, #software-engineering