Denoise · Twitter

The security focus is shifting from prompt injection in models to vulnerabilities in the agent orchestration layer, as noted by Anthropic, DeepMind, and @MalwareTechBlog.

Major AI labs like Anthropic and Google DeepMind are publicly releasing red-teaming frameworks and disclosures for vulnerabilities in autonomous agents.

• 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 emerging between terminal-first agents like Claude Code and IDE-centric tools like Cursor, with early adopters like @levelsio reporting faster shipping.

Anthropic released Claude Code 1.5, a terminal-native agent, intensifying the debate over whether IDEs or terminals are the future of coding.

• 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 convergence pattern is visible: OpenAI provides the SDK protocol, LangChain offers integration patterns, and Vercel/Replit supply the deployment runtimes for a complete agent stack.

OpenAI, Vercel, and Replit all launched new infrastructure for agent orchestration, tool calling, and deployment, signaling a maturing ecosystem.

• 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.

Mistral's release of a foundational OCR dataset indicates a strategic investment in improving the visual and text-extraction capabilities necessary for more advanced multimodal agents.

Mistral AI released a massive, cleaned 100M-row web OCR dataset for public use in training models.

• Mistral AI@MistralAIrising

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

Practitioners like @GregKamradt and @reach_vb are showing that large context doesn't solve retrieval; it creates new, harder problems like cache invalidation and structured memory management.

The conversation is shifting from RAG to more complex 'context engineering,' exploring failure modes in 10M-token windows and new 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 orchestration, previously a developer-focused concept from tools like Temporal, is now being integrated into general business software like Notion and Linear.

Workspace automation features are rolling out in mainstream productivity tools like Notion and Linear, automating triage and database updates.

• 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 approaches of @dotey (qualitative review) and Weights & Biases (quantitative benchmarking) show a field maturing from art to an engineering discipline.

Prompt engineering is evolving from sharing anecdotal tricks to systematic, large-scale benchmarking of system prompts against multiple 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's point, the bottleneck in training capable agents is shifting from compute access to sophisticated data filtering and curation pipelines.

A discussion emerged on the critical importance of curating synthetic data 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.