Denoise · Twitter

The security conversation has matured from prompt injection to the orchestration layer. Both Google DeepMind and Anthropic now frame the core risk in how agents interact with tools and each other.

Major labs and researchers are publishing formal frameworks and disclosures for red-teaming autonomous agents, focusing on complex interaction vulnerabilities.

• 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 battleground for AI coding assistants is moving from the IDE to the terminal. Early benchmarks from @swyx suggest significant performance divergence between models like Claude Code and older paradigms on complex tasks.

Anthropic's release of Claude Code 1.5, a terminal-native coding agent, is seen as a fundamental shift in developer experience, moving beyond IDE-based copilots.

• 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 on agent protocols and deployment standards is underway. OpenAI's SDK and LangChain's integration with Anthropic's MCP show competing and complementary efforts to define how agents interoperate.

OpenAI, Vercel, and Replit are all shipping infrastructure for deploying and orchestrating agents, signaling a rush to build the platform layer for this new primitive.

• 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 dominate the conversation, Mistral AI continues its distinct strategy of empowering the open-source community with foundational artifacts like large-scale, high-quality datasets.

Mistral AI released a large, cleaned, and licensed dataset for web optical character recognition (OCR) to support open model training.

• Mistral AI@MistralAIrising

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

As context windows expand dramatically, the bottleneck shifts from retrieval to memory management. Failures in cache invalidation (@reach_vb) and the need for structured memory (@mem0ai) highlight this new challenge.

Discussions have evolved from RAG to "context engineering," focusing on advanced memory systems and caching strategies for agents handling million-token contexts.

• 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 concept of autonomous, durable task execution is becoming a shared primitive. SaaS tools like Notion and infrastructure tools like Temporal are converging on this idea from opposite directions.

Workspace productivity tools like Notion and Linear are shipping agent-like automation features, while infrastructure players like Temporal position their platforms for agent orchestration.

• 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 of prompt optimization is industrializing. Efforts by @weights_biases to benchmark 40k variants show a move towards data-driven, empirical methods rather than relying on intuition.

Prompt engineering is shifting from anecdotal tips to systematic, large-scale benchmarking to find optimal system prompts across frontier 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 agent training matures, the challenge is no longer just generating synthetic data, but discerning its quality. @jerryjliu0 highlights the subtle but critical problem of data poisoning generalization.

The focus in agent training is on the data curation process, specifically filtering synthetic data that appears high-quality but degrades model generalization.

• Jerry Liu@jerryjliu0repeated

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