Denoise · Twitter

The security focus is shifting from the model itself to the agent's orchestration layer, where tool interactions and sandbox escapes, as highlighted by @GoogleDeepMind, are the new frontier.

Major AI labs are publishing formal frameworks and disclosures for red-teaming autonomous agents, moving beyond simple prompt attacks.

• 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.

@AnthropicAI's Claude Code 1.5 launch, supported by commentary from @karpathy and @swyx, reveals a concerted push to make the terminal the primary developer interface for AI.

Today's signal is dominated by the launch of terminal-native coding agents, with significant discussion on the workflow implications.

• 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 convergence is visible between protocol providers like @OpenAI and hosting platforms like @vercel and @replit, solidifying the 'durable agent worker' as a core cloud primitive.

The ecosystem is rapidly building the infrastructure for deploying agents, with major players releasing SDKs and hosting platforms.

• 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.

@MistralAI's contribution of a massive OCR dataset reinforces the idea that high-quality, specialized data remains a key differentiator, even as the application layer shifts to agents.

A single, large-scale open dataset for OCR was released, indicating continued investment in foundational data for vision tasks.

• Mistral AI@MistralAIrising

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

A tension is emerging between simple retrieval patterns (@llamaindex) and stateful, multi-layered memory systems proposed by @mem0ai and @GregKamradt, suggesting RAG is being unbundled.

The conversation shifts from basic RAG to more sophisticated 'context engineering' and complex memory architectures for stateful agents.

• 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 'auto-triage' feature from @linear and chained updates from @NotionHQ mirror the orchestration patterns in AI agent systems, showing a convergence of automation concepts across different domains.

Workspace automation tools from Notion and Linear are incorporating agent-like background processes for tasks like auto-triaging.

• 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.

Work by @weights_biases to map the efficient frontier for prompts demonstrates a maturing of the field, treating prompt engineering as a rigorous optimization problem, not an art.

The focus is on moving from manual prompt crafting to systematic, large-scale benchmarking to discover 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.

@jerryjliu0's insight on filtering synthetic data reveals that as models become more agentic, preventing 'poisoned generalization' becomes a core MLOps problem.

The main discussion point is the subtle but critical challenge of curating high-quality training data for agentic models.

• Jerry Liu@jerryjliu0repeated

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