Denoise · Twitter

The security focus is rapidly shifting from model-level prompt injection to vulnerabilities in agent orchestration and cross-tool communication.

Major AI labs like Anthropic and Google DeepMind are proactively releasing red teaming frameworks and disclosures for agent-based systems.

• 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 developer toolchain is bifurcating between traditional IDE extensions like Copilot and standalone terminal agents like Claude Code, with early adopters reporting productivity gains.

Anthropic's Claude Code 1.5 release defines a new product category of terminal-native agents, with benchmarks and early user reports validating its effectiveness.

• 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 full stack for agent deployment is rapidly emerging, with OpenAI defining protocols, Vercel providing serverless execution, and Temporal offering durable orchestration.

OpenAI, Vercel, and Replit released new infrastructure for deploying and orchestrating agents, from SDKs to edge runtimes.

• 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 agents dominate the conversation, foundational dataset work continues, with Mistral AI focusing on providing open, high-quality data for vision-language models.

Mistral AI released a large-scale, cleaned web OCR dataset to support the training of open multimodal models.

• Mistral AI@MistralAIrising

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

The conversation is evolving from RAG as a technique to memory as a system, with @mem0ai and @GregKamradt proposing layered memory models beyond simple vector search.

Developers are hitting the limits of simple RAG, leading to new frameworks for "context engineering" and more complex 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.

The agentic pattern is being embedded directly into existing SaaS platforms, automating user tasks without requiring users to build or manage standalone agents.

SaaS tools like Notion and Linear are launching features that automate internal team workflows, such as 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.

Prompt engineering is becoming more empirical, with firms like Weights & Biases applying systematic benchmarking to find optimal system prompts instead of relying on intuition.

Practitioners are sharing findings from large-scale system prompt benchmarking and reviews of production 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.

As agent capabilities grow, the bottleneck is shifting from raw data volume to the quality and filtration of synthetic data used for fine-tuning.

The focus in agent training data is on curating synthetic datasets to avoid models learning spurious correlations that harm generalization.

• Jerry Liu@jerryjliu0repeated

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