Denoise · Twitter

The security threat model has shifted from single-prompt exploits on models to multi-step attacks on the agent orchestration layer, a concern shared by Anthropic, Google DeepMind, and @AlexAlbert__.

Major AI labs are publishing formal red-teaming frameworks and disclosures for autonomous agents, focusing on complex jailbreaks and exploit chains.

• 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.

The primary developer interface is shifting from the IDE plugin (Copilot) to the standalone terminal agent (Claude Code), a structural change articulated by @karpathy and benchmarked by @swyx.

Anthropic's release of Claude Code 1.5, a terminal-native coding agent, dominates the conversation, with benchmarks and early adoption reports validating the trend.

• 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 clear convergence is happening around building the 'Kubernetes for agents,' with OpenAI, Vercel, and Replit all shipping primitives to standardize the agent deployment and orchestration stack.

Major infrastructure providers, including OpenAI, Vercel, and Replit, are releasing SDKs and deployment harnesses specifically for multi-worker 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 are the focus, foundational dataset releases like MistralAI's OCR data remain a critical and quiet enabler for the next generation of multimodal models that will power them.

MistralAI released a large-scale, 100M-row web OCR dataset for training vision models.

• Mistral AI@MistralAIrising

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

The vocabulary is shifting from 'RAG' to 'context engineering' (@GregKamradt). The problem is now framed as managing a complex memory hierarchy (@mem0ai), not just document retrieval.

The discussion is moving beyond simple RAG towards 'context engineering,' exploring long-context failure modes and more sophisticated memory architectures.

• 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.

Agent-like automation, prominent in developer tools, is now being integrated into general productivity software like Notion and Linear to handle routine knowledge work.

Workspace automation is a key theme outside of core AI development, with Notion and Linear launching features for auto-filling and auto-triaging.

• 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.

Prompt engineering is maturing from a craft into a science. Weights & Biases' large-scale benchmark represents a shift towards rigorous, data-driven optimization over individual 'tricks'.

The focus is on systematic, large-scale benchmarking of system prompts to discover an efficient frontier, moving beyond anecdotal prompt tricks.

• 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 agent capabilities grow, the bottleneck shifts back to data quality. The focus is now on sophisticated filtering of synthetic data, a crucial but often overlooked part of the training pipeline.

Jerry Liu shares insights on the critical process of curating high-quality synthetic data for training agents to avoid poisoning generalization.

• Jerry Liu@jerryjliu0repeated

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