Denoise · Twitter

The security conversation is moving up the stack from simple prompt injection to complex exploits in the agent orchestration layer, as noted by @MalwareTechBlog and @AlexAlbert__.

Red teaming efforts are now focused on autonomous agents, with major labs like Anthropic and DeepMind disclosing vulnerabilities and frameworks.

• 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 competition in coding agents, highlighted by @swyx's benchmarks, is now centered on terminal-based workflows and long-context reasoning, with developers like @levelsio already migrating.

Anthropic's launch of Claude Code 1.5, a terminal-native agent, validates the trend of moving AI coding tools out of the IDE.

• 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's a clear convergence on standardizing agent interaction, with OpenAI providing a protocol and LangChain showing how to adapt to it, indicating a move towards an interoperable agent ecosystem.

Major infrastructure providers like OpenAI, Vercel, and Replit are releasing SDKs and runtimes for orchestrating durable, multi-worker 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 the agent ecosystem is the focus, MistralAI's release shows that foundational dataset work continues to be a critical, albeit less visible, driver of progress for models that need to parse the visual web.

MistralAI released a large-scale, cleaned, and licensed web OCR dataset to support multimodal model training.

• Mistral AI@MistralAIrising

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

The simple RAG pattern is proving insufficient; actors like @mem0ai and @llamaindex are pushing towards multi-layered memory and knowledge graphs to handle complex reasoning.

Developers are exploring the failure modes of massive context windows and evolving RAG into more complex context engineering frameworks.

• 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 principles of autonomous orchestration from the AI agent world, like the durable workflows discussed by @temporalio, are being mirrored in SaaS productivity tools, suggesting a broader trend.

Workspace automation tools like Notion and Linear are releasing features for auto-triaging and chained updates, mirroring patterns 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 focus is shifting from anecdotal 'prompt tricks' (@dotey) to data-driven optimization, with platforms like Weights & Biases enabling systematic search for optimal system prompts.

Production-level prompt engineering is being systematized through large-scale benchmarking and the extraction of reusable patterns.

• 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.

@jerryjliu0 highlights a key challenge in the agent development lifecycle: filtering out synthetic data that appears useful but harms a model's ability to generalize.

The conversation centers on the crucial but difficult task of curating high-quality synthetic training data for agents.

• Jerry Liu@jerryjliu0repeated

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