Denoise · Twitter

The security frontier is shifting from prompt injection to agent tool-use exploits, with both @AnthropicAI and @GoogleDeepMind focusing on sandbox escapes and cross-tool data leakage.

Major labs are focusing on red-teaming autonomous agents and publicly disclosing vulnerabilities in their orchestration layers.

• 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 clear convergence is forming around the terminal as the primary AI coding interface, with @AnthropicAI's product launch validated by @karpathy's analysis and @levelsio's adoption.

The developer workflow is rapidly shifting towards terminal-native coding agents, led by Anthropic's new Claude Code 1.5.

• 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.

Actors like @OpenAI, @LangChainAI, @vercel, and @replit are converging on a common set of primitives for multi-worker orchestration, suggesting a protocol layer for agents is emerging.

The infrastructure for AI agents is standardizing around new SDKs and protocols for orchestration and deployment.

• 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.

It's a quiet day for this category, with @MistralAI's dataset release being a contribution to the community rather than part of a broader, active trend.

The only signal is a large-scale public dataset release for web OCR from MistralAI, aimed at open-source model training.

• Mistral AI@MistralAIrising

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

Thought leaders like @GregKamradt are explicitly declaring RAG insufficient, while tools like @mem0ai are building multi-layered memory systems, fragmenting the simple vector-retrieval consensus.

The conversation shifts from simple RAG to more sophisticated 'context engineering' and structured memory architectures for 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.

Workspace tools like @NotionHQ and @linear are converging on AI-driven workflow automation, mirroring the durable orchestration patterns seen in developer-focused tools from @temporalio.

General productivity tools are integrating agent-like automation for tasks like issue triage and database updates.

• 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 approach is shifting from artisanal prompt crafting (@dotey) to industrial-scale benchmarking (@weights_biases), treating prompt optimization as a formal search problem.

Prompt engineering is becoming more systematic, with a focus on 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 highlights a core challenge in scaling agents: filtering synthetic data to avoid poisoning generalization, a problem that is fundamental to the entire MLOps for agents stack.

Discussion centers on the critical and difficult task of curating high-quality datasets for training capable agents.

• Jerry Liu@jerryjliu0repeated

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