Denoise · Twitter

The focus is shifting from simple prompt injection to complex orchestration-layer exploits, as detailed in releases from @AnthropicAI and @GoogleDeepMind.

Frontier labs and security researchers are proactively releasing frameworks and disclosures for agent-specific vulnerabilities like jailbreak chains and cross-tool leakage.

• 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 release of an integrated agent product by @AnthropicAI, validated by @karpathy, directly competes with the more generalized tool-and-protocol approach from OpenAI.

Anthropic's release of Claude Code 1.5 has established the terminal as a primary interface for AI agents, with immediate community validation and benchmarking.

• 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 de facto agent stack is materializing where platforms like @OpenAI define protocols while @vercel and @replit provide the execution environments, requiring developer integration.

Major infrastructure providers like OpenAI, Vercel, and Replit are shipping agent-specific SDKs, protocols, and deployment runtimes in parallel.

• 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 narrative dominates, @MistralAI's contribution of a foundational dataset represents a differentiated strategy focused on open model inputs rather than tooling.

Mistral AI released a large, open dataset for web OCR, focusing on foundational data enablement for vision models.

• Mistral AI@MistralAIrising

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

A fault line is emerging between scaling context windows (@reach_vb) and building explicit, structured memory systems (@mem0ai, @GregKamradt), suggesting no single architecture has won.

The conversation is shifting from simple RAG to more complex "context engineering" and dedicated agent memory layers with distinct working and long-term stores.

• 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 principles of durable workflows from @temporalio are converging with the stateful, resumable needs of AI agents, indicating a consolidation of the two domains.

Workspace automation tools like Notion and Linear are shipping agent-like features, while orchestration engines like Temporal are positioning for agent workloads.

• 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 work by @weights_biases to map the efficient frontier for system prompts signals a maturation of prompt engineering into a more rigorous, data-driven discipline.

Efforts are underway to move beyond anecdotal prompt advice to systematic, large-scale benchmarking of system prompts across 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.

@jerryjliu0's focus on synthetic data quality for agents highlights a critical bottleneck that is less about raw compute and more about high-quality training signals.

A key challenge in training capable agents is identified as the curation and filtering of synthetic datasets to avoid generalization poisoning.

• Jerry Liu@jerryjliu0repeated

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