Open Models, Agent Infra, and the Search for Self-Sufficiency

Research

New Research on Making LLMs Their Own Search Algorithm

This paper introduces an 'Algorithm of Thoughts' that teaches models to search for solutions using in-context learning, essentially turning the LLM into its own search algorithm. For builders, this means potentially fewer external tool calls and more autonomous problem-solving for complex tasks.

News

Kimi K2 Goes Open-Source, Boosting the Open Agent Stack

Kimi K2's open weights release signals a major shift toward powerful, openly available models that can form the backbone of autonomous agent systems. This gives builders a high-performance, accessible foundation for building and fine-tuning their own specialized agents.

The New Fog of AI Agents in Business Operations

The diffusion of AI agents into critical infrastructure is creating a new kind of uncertainty, where decision-making becomes opaque and accountability blurs. This is the foundational operational challenge for any organization deploying autonomous systems at scale.

Tools

Firecrawl Launches Agent-First Endpoints for Web Data

Firecrawl's new endpoints are built specifically for autonomous agents, handling the messy parts of web interaction like waiting for JavaScript to render and extracting clean, structured data. This is critical infrastructure for any agent that needs to reliably interact with the live web at scale.

Analysis

Ethan Mollick on the New Reality of AI Coding Coworkers

Mollick unpacks the reality of working alongside AI coding agents, highlighting both their impressive productivity boosts and the new coordination challenges they introduce. The piece is a must-read for understanding how team dynamics and development workflows are being fundamentally rewritten.