Denoise · Twitter

The security frontier is shifting from model-level prompt injections to system-level vulnerabilities in agent orchestration, a concern echoed by Anthropic, Google DeepMind, and AlexAlbert__.

The focus is on establishing security best practices for AI agents, with major labs releasing red-teaming frameworks and vulnerability disclosures.

• Anthropic@AnthropicAIrising

  Responsible disclosure on a Claude jailbreak chain we patched last week. Full write-up including our red team timeline.

• Google DeepMind@GoogleDeepMindrising

  New red team framework for prompt injection in autonomous agents. Covers cross-tool leakage, scanner evasion, and sandbox escape patterns.

• MalwareTech@MalwareTechBlogrepeated

  Autonomous agent running pentest flows against a real SaaS. First real-world run: fewer false positives than I expected on the vulnerability surface.

A clear competition is forming between integrated IDE assistants like Copilot and standalone terminal agents like Claude Code, with developers like swyx and levelsio providing early benchmarks and adoption signals.

Anthropic's release of Claude Code 1.5, a terminal-based agent, dominates the conversation, sparking debate on the future of developer workflows.

• Anthropic@AnthropicAIrising

  Claude Code 1.5 is live. Terminal-native coding agent with full Claude Opus reasoning, file-ops sandbox, and session replay.

• Andrej Karpathy@karpathyrising

  The developer-experience shift from IDE to terminal agent is underrated. Coding workflows are about to look nothing like 2024.

• swyx@swyxrising

  Codex vs Claude Code terminal agent benchmarks. Pass@1 diverges more than I expected on the long-context editor tasks.

• DSPy@dspy_airising

  DSPy 3.0: prompt optimization via compile-time search over system prompt variations. Benchmarks inside.

• @levelsio@levelsiorising

  Switched my whole editor setup to Claude Code this week. Shipping faster than when I used Cursor + Copilot.

A standardized 'agent cloud' layer is emerging, with OpenAI defining protocols, platforms like Vercel and Replit providing the execution environment, and LangChain acting as a bridge for existing systems.

Major platforms like OpenAI, Vercel, and Replit are rolling out SDKs and runtimes specifically for deploying and managing autonomous agents.

• OpenAI@OpenAIrising

  New agent SDK: protocol-level tool calling, deployment harness, and multi-worker orchestration primitives. Docs live.

• LangChain@LangChainAIrising

  MCP protocol integration thread. How to wire existing LangGraph agents into the Anthropic Model Context Protocol server spec.

• Vercel@vercelrising

  Edge runtime for agent workers is live. Spawn durable background agents from any serverless deployment.

• Alex Albert@AlexAlbert__rising

  When your security scanner finds nothing scary on an agent deploy, check the orchestration layer again. That's usually where the jailbreak sneaks through.

• Replit@replitrising

  New agent deployment harness. One command to go from local orchestration to hosted agent worker.

While agents receive most of the attention, foundational dataset contributions from players like MistralAI remain a critical, albeit less visible, driver of progress in multimodal AI.

MistralAI released a large, clean web OCR dataset, providing a foundational resource for training next-generation multimodal models.

• Mistral AI@MistralAIrising

  Open dataset release: 100M-row web OCR dataset. Cleaned, licensed, ready to train.

A split is appearing between just-in-time retrieval (RAG) and persistent, structured memory (mem0ai), with practitioners like GregKamradt arguing for new frameworks to manage massive context windows.

The discussion is evolving from simple RAG implementations to more complex 'context engineering' frameworks for agent memory and retrieval.

• Vaibhav Srivastav@reach_vbrising

  Tested the new 10M context memory window end to end. Surprising failure modes around rag retrieval cache invalidation, thread below.

• Greg Kamradt@GregKamradtrising

  RAG is dead, long live context engineering. My framework for when to cache, when to retrieve, and when to just dump memory into the prompt.

• mem0@mem0airising

  Memory layer for agents: differentiating working memory from the subconscious store. Vector index isn't enough anymore.

• LlamaIndex@llamaindexrepeated

  Knowledge graph retrieval walkthrough: when semantic vector search misses, graph hop beats it every time.

The autonomous task management seen in Notion and Linear's new features reflects a broader trend of embedding agent-like logic directly into business software, paralleling the goals of more general AI agents.

Workspace automation features are being released by SaaS tools like Notion and Linear, automating triage and data management.

• Notion@NotionHQrising

  Notion workspace automation is out of beta. Auto-fill tables, chained updates across databases, and a new audit log surface.

• Linear@linearrising

  Linear now auto-triages incoming issues. Quiet launch, but already our favorite workspace feature of the year.

• Temporal@temporaliorepeated

  Orchestrating agents with durable workflows: replayable, resumable, and multi-worker by default. Walkthrough from our infra team.

• James Clear@jamesclearrepeated

  The best habit tracker is the one you actually open. Three open-source alternatives worth trying.

The discipline of prompt engineering is professionalizing, with platforms like Weights & Biases enabling rigorous, data-driven optimization, treating prompts as a core tunable parameter in ML systems.

The focus is on moving from anecdotal prompt tricks to systematic, large-scale benchmarking of system prompts across different models.

• dotey@doteyrising

  Five prompt tricks learned this week from reviewing 200 production prompts. Short thread.

• Weights & Biases@weights_biasesrising

  System prompt benchmarking at scale: we ran 40k variants across 6 frontier models. The efficient frontier is not where you think.

As per jerryjliu0, the bottleneck for advanced agents is shifting from model architecture to data quality, specifically the difficult task of filtering out synthetic data that harms generalization.

A key challenge highlighted is the curation of synthetic datasets for agent training to avoid performance degradation.

• Jerry Liu@jerryjliu0repeated

  Dataset curation for agent training: how we filter synthetic data that looks good but poisons generalization.