Denoise · Twitter

The primary security risk is shifting from the LLM itself to the agent's orchestration layer, where vulnerabilities like cross-tool leakage and sandbox escapes emerge.

Red team efforts from Anthropic and DeepMind are now focused on the novel attack surfaces of autonomous agents, beyond simple prompt injection.

• 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 split is visible: Anthropic is betting on a terminal-centric agent workflow, while tools like Cursor and Copilot remain integrated with the IDE, creating a new competitive front.

Anthropic's release of Claude Code 1.5, a terminal-native agent, sparks debate on a workflow shift away from traditional IDEs.

• 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.

There is a rapid convergence on the core components for agents: a standardized tool-calling protocol (OpenAI/Anthropic), an orchestration layer (LangChain), and a serverless deployment target (Vercel/Replit).

OpenAI, Vercel, and Replit released new SDKs, runtimes, and deployment harnesses, signaling a push to standardize and host AI 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 agent orchestration dominates discourse, the quiet release of foundational datasets like this one for OCR indicates that building more sensorily rich models remains a core, parallel effort.

MistralAI released a large, 100M-row web OCR dataset, providing a foundational asset for training multimodal models.

• Mistral AI@MistralAIrising

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

A consensus is forming that vector search alone is insufficient for agents. Frameworks like LlamaIndex and mem0.ai are proposing hybrid approaches combining vector stores with knowledge graphs and structured memory.

With 10M token context windows, discussion shifts from simple RAG to 'context engineering'—complex strategies for memory, caching, 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 durable workflow pattern, articulated by Temporal, is becoming a common paradigm for both enterprise SaaS automation (Notion) and complex, multi-worker AI agent systems.

Workspace automation tools like Notion and Linear are releasing features that parallel the orchestration logic seen in AI agents.

• 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 practice is maturing from craft to science, with organizations like Weights & Biases running tens of thousands of prompt variations to find optimal performance, treating it as a formal tuning problem.

The focus in prompt engineering is shifting from anecdotal tricks to large-scale, systematic benchmarking of system prompts.

• 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.

The problem has evolved beyond mere data generation; as @jerryjliu0 notes, the focus is on developing sophisticated filtering techniques to prevent 'data poisoning' from plausible but incorrect synthetic examples.

The key challenge in data for agents is now curating high-quality synthetic data and filtering out examples that can harm generalization.

• Jerry Liu@jerryjliu0repeated

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