Denoise · Twitter

The security conversation is moving from model-level prompt injection to system-level agent orchestration vulnerabilities, with Anthropic and Google DeepMind leading formal research.

Major AI labs are publicly disclosing red-teaming frameworks and patched jailbreaks for autonomous agents, while independent researchers are testing them in real-world pentesting.

• 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 battle for developer adoption is moving to the terminal. Claude Code, alongside frameworks like DSPy, are creating new developer experiences that abstract away traditional prompting.

Anthropic's Claude Code 1.5 launch dominates the conversation, framing the shift from IDE-based copilots to full terminal-native coding agents.

• 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 consensus is forming around durable, multi-worker agents as a core primitive, with Vercel, Replit, and OpenAI all releasing competing but conceptually similar deployment tools.

The focus is on standardizing agent orchestration, with OpenAI releasing an SDK and frameworks like LangChain adapting to new protocols like MCP.

• 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 the agent conversation is text-focused, Mistral AI's dataset release signals continued investment in foundational data for vision, a key component for future multimodal agents.

Mistral AI released a large-scale, cleaned web OCR dataset for public use, contributing a key resource for multimodal model training.

• Mistral AI@MistralAIrising

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

Developers like @GregKamradt and tools like mem0.ai are arguing that simple vector retrieval (RAG) is insufficient, pushing towards structured memory systems for agents.

The conversation is shifting from large context windows to sophisticated 'context engineering,' including caching strategies, memory hierarchies, and graph 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.

The concept of autonomous, orchestrated workflows is expanding beyond code to general business productivity, with Notion and Linear embedding agent-like behaviors directly into their platforms.

Productivity tools like Notion and Linear are releasing beta features for workspace automation and auto-triaging, mirroring the agentic patterns seen in development tools.

• 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 focus is shifting from simple prompt 'tricks' to scalable, benchmark-driven optimization, with players like Weights & Biases providing data on the non-obvious performance of system prompts.

Practitioners are sharing hard-won knowledge on production prompt engineering and the surprisingly complex landscape of system prompt performance across 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.

As agent training becomes more common, the infrastructure problem shifts from raw compute to the subtle art of dataset filtering to avoid generalization failures, as highlighted by @jerryjliu0.

The focus is on the data curation challenges specific to training agents, particularly filtering synthetic data that can harm generalization.

• Jerry Liu@jerryjliu0repeated

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