Denoise · Twitter

The security focus is shifting from prompt injection in chat models to complex, multi-step exploits in agent orchestration layers, a concern echoed by @AnthropicAI, @GoogleDeepMind, and @AlexAlbert__.

Major labs are publicly disclosing agent jailbreaks and releasing formal red teaming frameworks for testing autonomous agent 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 conversation, led by voices like @karpathy and @swyx, is converging on the idea that standalone terminal agents are poised to replace IDE-integrated tools like Copilot and Cursor.

Anthropic's release of Claude Code 1.5, a terminal-native agent, has triggered a wave of adoption and comparative benchmarking against existing tools.

• 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 clear convergence is happening in agent orchestration, with OpenAI's SDK, LangChain's protocol integration, and Temporal's durable workflows all providing primitives for managing multi-worker agents.

OpenAI, Vercel, and Replit all released infrastructure for building and deploying agents, focusing on orchestration, tool-calling, and serverless 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.

This large-scale, cleaned data release by Mistral AI signals a continued strategy to empower the open-source community to build competitive specialized models.

Mistral AI released a 100M-row web OCR dataset, providing a foundational resource for training open-source models on document understanding tasks.

• Mistral AI@MistralAIrising

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

Practitioners like @GregKamradt are explicitly framing this shift as "context engineering," while @reach_vb and @mem0ai demonstrate the need for sophisticated memory architectures beyond simple vector search.

The conversation around context is maturing from basic RAG to complex strategies for caching, retrieval, and layered memory in multi-million token windows.

• 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 pattern of "agent-ification" is expanding from developer tools to general business software, with Notion and Linear automating complex, chained user tasks.

Workspace automation features are being released by Notion and Linear, applying agent-like autonomous capabilities to productivity software.

• 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.

This signals a maturation of the field, where large-scale, data-driven approaches by firms like Weights & Biases are now complementing practitioner heuristics from individuals like @dotey.

The practice of prompt engineering is moving from anecdotal tricks to systematic, large-scale benchmarking to find optimal 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.

As noted by @jerryjliu0, the key challenge is no longer data quantity but data quality, specifically identifying and removing synthetic data that could "poison" the agent's performance.

The focus in data preparation for agent training is on sophisticated filtering of synthetic data to improve model generalization.

• Jerry Liu@jerryjliu0repeated

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