Denoise · Twitter

The conversation is maturing from prompt injection to complex orchestration-layer attacks, with @GoogleDeepMind and @MalwareTechBlog defining this new surface.

The focus shifts to agent-specific vulnerabilities, with a major disclosure from Anthropic and new red teaming frameworks for multi-step 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 release, amplified by @karpathy's commentary, directly challenges the established Copilot/IDE model of developer assistance.

Major releases center on terminal-native coding agents, signaling a potential paradigm shift away from IDE-centric workflows.

• 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 between @OpenAI's SDK and runtime providers like @vercel and @replit reveals a shared vision of agents as deployable, stateful services.

The ecosystem is rapidly building orchestration and deployment layers for agents, moving from APIs to durable, hosted workers.

• 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 agent orchestration dominates today's discourse, @MistralAI's data release is a reminder that model capability is still gated by high-quality, large-scale data.

A single, significant open dataset release for web OCR from MistralAI provides foundational resources for training multimodal agents.

• Mistral AI@MistralAIrising

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

Practitioners like @GregKamradt and startups like @mem0ai are moving beyond vector search, reframing the problem as managing an agent's working vs. long-term memory.

The discussion is evolving from simple RAG to 'context engineering,' focusing on sophisticated 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.

Products from @NotionHQ and @linear show how agentic patterns are being integrated into mainstream SaaS, moving beyond developer-focused tools.

Workspace automation tools are embedding agent-like capabilities for tasks like issue triage and database updates, abstracting AI complexity.

• 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.

Despite the rise of high-level frameworks, work by @weights_biases shows that low-level system prompt optimization remains a key performance differentiator.

Practitioners continue to share tactical prompt engineering advice and scale up system prompt benchmarking across frontier 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.

@jerryjliu0's note on data filtering reveals a key bottleneck in scaling agent capabilities: avoiding dataset poisoning that harms generalization.

The conversation highlights the critical importance of curating high-quality synthetic data for training effective and generalizable agents.

• Jerry Liu@jerryjliu0repeated

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