Alex Mashrabov, CEO of Higgsfield, explains how agentic systems choose among models and provision their own GPU infrastructure to deliver consistent, multi-step creative media. He covers the trade-offs between speed, controllability, and dedicated deployment, and shares the 10-hour AI-generated live stream as an example of what becomes possible for small teams.
Jan Čurn, founder and CEO of Apify, argues that the backlash against MCP targets the wrong layer. Context bloat and token burn come from naive agent harnesses, not the protocol. He reviews the industry's three main fixes (sub-agents, progressive tool discovery and Code Mode), explains why agents work well with CLIs, proposes MCP for remote access and CLIs for local use, and demos mcpc, Apify's open-source universal MCP CLI client, alongside early Connector Evals results.
Episode 1022 dissects the crucial aspect of effectively steering AI agents by determining the optimal placement of instructions to ensure reliability and cost-efficiency. It explores seven distinct methods for instruction delivery, contrasting instructions as probabilities with hooks as guarantees, and highlights the industry-wide adoption of standards like `agents.md` and the importance of human-crafted guidance for superior agent performance.
Alistair Pullen, CEO of Cosine, discusses the UK's sovereign AI initiative, born from US export controls. He outlines Cosine's unique economic model, competing with "millions" against "billions" by licensing models instead of hosting inference. Pullen delves into why open models lag frontier systems, emphasizing active parameters and post-training data. He explains Cosine's innovative approach to "slop" through process-based RL and credit attribution, advocating for runtime proof in code review. The conversation covers their hierarchical "Swarm" sub-agent system, the challenges of memory, and advanced synthetic data generation, concluding on the geopolitical impact of export controls as an unexpected catalyst for UK AI.
Aditya Bhargava argues that the 'harness' (the surrounding logic and tools) is more critical than the underlying LLM model itself for building effective AI agents. He proposes that focusing on sophisticated harness design, supported by his new language Agency, can enable local, open-source models to achieve performance comparable to large proprietary models, thereby reducing dependency and fostering innovation.
The Arize team shares lessons from building their AI agent, Alyx, which analyzes its own trace data. They detail their journey from failed attempts like naive truncation and summarization to a successful strategy combining head/tail preservation with a retrievable memory store and using sub-agents to manage context complexity.