Denoise · Twitter

The focus on agent security is shifting from theoretical prompt injection to practical, automated pentesting and orchestration-level vulnerabilities.

Major labs like Anthropic and Google DeepMind are releasing formal red-teaming frameworks and disclosures for agent security.

• 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 is now centered on terminal-native agents, with Anthropic's Claude Code directly challenging the Copilot/IDE paradigm, and DSPy focusing on optimizing the underlying prompt logic.

Anthropic's launch of Claude Code 1.5, a terminal-native agent, prompts discussion on the 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.

A clear convergence pattern is emerging around agent orchestration, with platforms like OpenAI, Vercel, and Replit building hosting layers while LangChain focuses on protocol interoperability.

OpenAI, Vercel, and Replit released new infrastructure for deploying and orchestrating agents, signaling a race to own the agent runtime.

• 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.

The primary contribution in this space today is a public dataset from Mistral AI, indicating that foundational data curation is still a key bottleneck.

Mistral AI released a large-scale, 100M-row web OCR dataset for training multimodal models.

• Mistral AI@MistralAIrising

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

A fault-line is appearing between simply expanding context windows (per @reach_vb's failure modes) and building explicit memory layers (@mem0ai, @GregKamradt) for stateful agents.

The conversation is moving from "RAG vs. long context" to "context engineering," with new frameworks for managing memory 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.

While SaaS tools like Notion and Linear build specific automations, infrastructure like Temporal provides generalized primitives for the same class of problems, particularly for AI agents.

Workspace automation features are launching in Notion and Linear, while Temporal highlights its use case for orchestrating 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 field is bifurcating between artisanal prompt crafting (@dotey) and industrial-scale, data-driven optimization of system prompts (@weights_biases, @dspy_ai).

Practitioners are sharing hands-on prompting techniques while Weights & Biases provides large-scale benchmark data on system prompt performance.

• 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 discussion highlights a critical, often overlooked step in the MLOps lifecycle for agents: sophisticated data curation beyond simple generation.

Jerry Liu shares insights on the necessity of filtering synthetic data to avoid poisoning model generalization during agent training.

• Jerry Liu@jerryjliu0repeated

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