Denoise · Twitter

The attack surface has officially moved up the stack from prompt injection to agent orchestration, cross-tool leakage, and sandbox escapes.

Major labs like Anthropic and Google DeepMind are publicly disclosing red-teaming frameworks for autonomous agents, focusing on systemic 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.

A direct competition is forming between Anthropic's terminal-centric agent and the established IDE-integrated model of GitHub Copilot, with early adopters reporting productivity gains.

Anthropic's release of Claude Code 1.5, a terminal-native agent, sparks discussion on the fundamental shift in developer workflows 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.

The race is on to provide the 'Heroku for agents,' with protocol-level standards (OpenAI SDK) and managed runtimes (Vercel, Replit) competing for developer adoption.

OpenAI, Vercel, and Replit all launched new infrastructure for deploying, hosting, and orchestrating agents, indicating a rush to build platform-level support.

• 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 most of the day's activity is on the agent application layer, Mistral's move is a reminder that progress still depends on foundational data curation.

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

• Mistral AI@MistralAIrising

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

The term 'RAG' is being replaced by more nuanced frameworks like 'context engineering' (GregKamradt) and layered memory systems (mem0ai) to cope with architectural complexity.

Discussions center on managing new failure modes in massive 10M+ token context windows, moving beyond simple retrieval augmentation.

• 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 trend shows a convergence where general-purpose agents (from AI labs) and specialized, embedded automations (from SaaS vendors) are competing for the same user workflows.

SaaS platforms like Notion and Linear are shipping embedded, agent-like automation features for workspace and issue management.

• 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.

Weights & Biases' 40k-variant experiment exemplifies the industrialization of prompt optimization, treating it as a formal hyperparameter tuning problem rather than an art.

The focus in prompt engineering is shifting from anecdotal tricks to systematic, large-scale benchmarking of system prompts against multiple 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.

Jerry Liu's point on filtering synthetic data reveals a key challenge: preventing data that looks clean but poisons model generalization is crucial for building reliable agents.

The conversation highlights the critical, behind-the-scenes work of curating high-quality datasets for training performant agents.

• Jerry Liu@jerryjliu0repeated

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