Parse notes
Sponsored
Most agents re-decide their next move at every step, so every run is non-deterministic end to end: hard to debug, hard to govern and hard to trust. This webinar breaks down a different approach: separating the agent's reasoning from how that reasoning gets executed.
Prime Intellect's verifiers v1 is an overhaul of its environment stack for the modern era of agentic RL and evals. It decomposes environments into a task set, a harness, and a runtime. verifiers v1 enables complex agentic tasks like coding and computer use at scale in any harness.
Sakana AI's research on collective intelligence has extended into physical hardware with "Smart Cellular Bricks." These bricks employ local communication and neural networks to autonomously classify and reconstruct 3D shapes without centralized control. In experiments, the system achieved high accuracy in shape classification and demonstrated robustness against noise and module failures.
This post looks at the data that xAI's official Grok Build coding CLI transmits and stores. The CLI transmits the contents of the files it reads to xAI verbatim and unredacted. It uploads whole repositories independent of what the agent reads. The storage destination is a Google Cloud Storage bucket. There's no proof that xAI trains on the data.
Cognition replaced Opus 4.8 with Fable 5, and Devin's bill went down. Fable 5 costs twice as much per token as Opus 4.8, but Cognition created an architecture that resulted in Fable costing less while scoring higher. This post looks at how the company achieved this and what that means for pricing agentic work.
Sponsored
Long-lived agent credentials are a breach waiting to happen. This workshop [12] covers agent identity and access management on AWS, including scoped credentials via AWS STS, automatic rotation with AWS Secrets Manager, and audit trails via AWS CloudTrail. Plus technical demos of Auth0 and CyberArk PAM.
Microsoft described the infrastructure and key mechanisms needed to operate AI agents across Foundry and its large-scale Copilot products. Key practices included treating retrieval as a sub-agent, assigning agents distinct identities and workspaces, and using rubric-based evaluations with automated improvement loops.
Mantis Skill is a decoupled, sequential, and security-focused set of Skills designed for use with Coding Agents. It is intended to be a flexible foundation that users can adapt, tune, and extend, rather than a rigid set of instructions. Google recommends using AI to iterate on the skills and use internal documentation, coding standards, and build systems to augment the threat model. It also strongly recommends adapting risk calibration to the environment and risk tolerance.
DeepMind researchers introduced GenCeption, which repurposes a pretrained video generation model as a unified vision system controlled through text instructions.
Long-Horizon Terminal-Bench evaluates whether LLM agents can sustain productive work across hundreds of terminal interactions. Its 46 stateful tasks use hidden verifiers that rebuild and inspect final artifacts rather than trusting an agent's reported progress.
Zvi Mowshowitz argued GPT-5.6 Sol delivers the strongest balance of reasoning quality, speed, and cost among current frontier models, making it the default choice for demanding knowledge work. He highlights its long-horizon execution, effective computer use, and agentic workflows while noting that task-specific model selection still matters for peak performance.
AI is emerging as a transformative tool, empowering individuals to create and innovate, rather than just consume. It reduces the barriers to production, allowing people without technical backgrounds to build practical and creative applications. Unlike consumption-focused technologies, AI nurtures individuality and creativity, fostering a culture where personal projects and ideas can flourish and become meaningful contributions.
Sponsored
Based on benchmarking tests, CData Connect AI can cut LLM context handling costs by up to 97.6% when using sources such as Salesforce,
Vercel reported that open-weight models handled 29% of AI Gateway token volume in June while accounting for less than 4% of spending.
He will be working with Tom Brown on the compute team.
Manus Auto-Publish allows automatic deployment of successful builds to live URLs without manual intervention.
Intelligence and the economy are converging into a single new system.