Denoise · Twitter

The focus of AI security is shifting from model-level vulnerabilities to the agent orchestration layer, where cross-tool leakage and sandbox escapes are the new attack vectors.

Major labs like Anthropic and Google DeepMind are releasing formal frameworks and disclosures for agent-specific security threats like jailbreaks and 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 consensus is forming around the terminal as the primary interface for developer agents, a shift articulated by @karpathy and implemented by Anthropic, challenging the dominance of IDE-based tools.

Anthropic's release of Claude Code 1.5 as a terminal-native agent, alongside benchmarks from @swyx and user reports, solidifies the move toward agent-centric development workflows.

• 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.

Model providers (OpenAI), frameworks (LangChain), and cloud platforms (Vercel) are converging on a common set of primitives for agent orchestration and deployment, solidifying a new infrastructure layer.

A wave of new infrastructure for agents has been released, including an SDK from OpenAI and deployment harnesses from Vercel and Replit for orchestrating agent workers.

• 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 agentic workflows dominate the discourse, foundational work on high-quality, large-scale datasets like Mistral's OCR corpus remains a key driver for future multimodal capabilities.

Mistral AI released a large-scale, 100M-row open dataset for web OCR, providing a significant resource for training new multimodal models.

• Mistral AI@MistralAIrising

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

Established RAG proponents like @GregKamradt and frameworks like LlamaIndex are now exploring more nuanced approaches, differentiating between working memory, long-term storage, and knowledge graphs.

With 10M context windows being tested, the conversation is shifting from simple RAG to more complex 'context engineering' and tiered memory systems.

• 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.

A convergence is happening between business automation in SaaS (Notion, Linear) and agent orchestration in dev tools (Temporal), with both adopting durable, multi-step workflow patterns.

Workspace automation features from Notion and Linear are now generally available, alongside discussion on using durable execution frameworks like Temporal for 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 engineering is maturing, with firms like Weights & Biases applying rigorous, data-driven methods to find optimal system prompts, moving beyond individual developer heuristics.

The focus on system prompts is moving from anecdotal tricks to large-scale, systematic benchmarking, as shown by a 40k-variant study from Weights & Biases.

• 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 advance, the bottleneck shifts to subtle data quality issues. @jerryjliu0 highlights the emerging problem of filtering synthetic data that appears valid but poisons model generalization.

The discussion around data for training agents is getting more specific, focusing on nuanced filtering techniques for synthetic data to avoid performance degradation.

• Jerry Liu@jerryjliu0repeated

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