Agentic AI Daily

THE DAILY READING LIST

Less noise.
More agency.

Original perspectives on AI agents, tools, and the systems that make them useful. Up to ten worthwhile reads, collected in the cloud.

2026-11-1110 readsPast 7 days · no padding
  1. Microsoft Research

    Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents

    Source excerptMicrosoft researchers have discovered “Lazarus,” a backdoor technique that can reprogram dormant neurons in an AI model, causing an otherwise normal AI agent to execute malicious commands when a hidden trigger appears. The threat is especially serious for AI agents with tool or terminal access. A triggered model could quietly download and run attacker code, […] The post Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents...

    Why this topic matters · rule-basedRelevant to AI systems that take actions, including coding and computer-use workflows.

    AI agents
  3. LangChain Older pick · over 48 hours

    What Is Eval-Driven Development? How to Improve AI Agents

    Source excerptEval-driven development helps you build AI agents against a consistent set of metrics. Define good behavior, write the eval first, then compare experiments.

    Why this topic matters · rule-basedRelevant to AI systems that take actions, including coding and computer-use workflows.

    AI agents
  4. OpenAI Older pick · over 48 hours

    Measuring how AI agents handle everyday wet-lab science

    Source excerptOpenAI introduces BioBasicsBench, a 100-task benchmark measuring how well AI agents interpret raw experimental data and handle everyday wet-lab science.

    Why this topic matters · rule-basedRelevant to AI systems that take actions, including coding and computer-use workflows.

    AI agents
  5. Microsoft Research Older pick · over 48 hours

    Formally verifying agent-written full-stack web apps with PDD Kit

    Source excerptAI agents can generate code at a scale that’s hard for humans to review line-by-line. PDD Kit helps ensure the code is doing what the developer intended by creating formal specifications and testing against them. It converts developer intent into a Verus specification, proves AI-generated code meets it for every input, and uses tests where proofs aren't practical. Developers review intent, not every line. The post Formally verifying...

    Why this topic matters · rule-basedRelevant to AI systems that take actions, including coding and computer-use workflows.

    AI agents
  6. OpenAI Older pick · over 48 hours

    CodeRabbit scales agentic software development and ships 20x more

    Source excerptCodeRabbit uses OpenAI models to govern and trust AI-generated code and Codex to ship 20 times more software with the same engineering team.

    Why this topic matters · rule-basedRelevant to designing systems that plan and act across multiple steps.

    Agentic systems
  7. Simon Willison Older pick · over 48 hours

    Hide your slop PR descriptions behind <details><summary>

    Source excerptMost of the changes I make to my projects these days come in the form of PRs opened for me by Claude Code or Codex. The PR is my unit of work - I use them to bundle commits together, iterating on them until a feature is complete, document decisions and then merge the finished result. Today's coding agents are great at writing PR descriptions. They're too good at describing the work The problem is that they are too good. An agent-written PR...

    Why this topic matters · rule-basedRelevant to AI systems that take actions, including coding and computer-use workflows.

    AI agents
  8. Anthropic Older pick · over 48 hours

    The components of agent memory

    Source excerptFor Claude to learn from experience, it has to decide what’s worth remembering before it knows what’s next. Here’s how “dreaming” helps it organize memory and correct mistakes.

    Why this topic matters · rule-basedRelevant to implementing agents; consult the original for capabilities and limitations.

    Agent engineering
  9. LangChain

    How our GTM Agent builds consistent, high-quality sales artifacts

    Source excerptLearn how LangChain built GTM Agent on Deep Agents to build high-quality, custom sales artifacts trained on business context.

    Why this topic matters · rule-basedRelevant to implementing agents; consult the original for capabilities and limitations.

    Agent engineering
  10. LangChain Older pick · over 48 hours

    Agents Are Going Async

    Source excerptAsync subagents let an agent stay responsive while background workers run long tasks. How the orchestrator, notifications and approvals work in Deep Agents.

    Why this topic matters · rule-basedRelevant to implementing agents; consult the original for capabilities and limitations.

    Agent engineering

A reading list, not an AI news summary.

Excerpts are shortened publisher feed or page descriptions, not independently verified reporting or AI-written summaries. Topic context is a rule-based explanation, not a claim about an article's findings. Articles must have a publication date within seven days and match agent-related terms. Relevance, freshness, and source variety determine the order. Picks from the previous 14 days are excluded. Articles older than 48 hours are labeled. No older backfill is used.

Source coverage & access notes
  • OpenAI
    150 dated entries parsed

    0 entries skipped (missing metadata or outside first-party scope)

  • Anthropic
    12 dated entries parsed

    Inspected 12 index links in publisher order

  • Google DeepMind
    100 dated entries parsed

    0 entries skipped (missing metadata or outside first-party scope)

    Article access failed: https://deepmind.google/blog/gemini-4-carbon-is-now-available: Cross-publisher or non-HTTPS redirect refused

    Article access failed: https://deepmind.google/blog/gemini-omni-12-flash-fast-and-cost-effective-with-longer-clips: Cross-publisher or non-HTTPS redirect refused

    31 dated entries have no usable excerpt and cannot be selected

  • Microsoft Research
    10 dated entries parsed

    0 entries skipped (missing metadata or outside first-party scope)

  • Hugging Face
    87 dated entries parsed

    63 entries skipped (missing metadata or outside first-party scope)

    82 dated entries have no usable excerpt and cannot be selected

  • LangChain
    12 dated entries parsed

    Inspected 12 index links in publisher order

  • LlamaIndex
    Unavailable: No dated articles found; publisher metadata may be unavailable

    Skipped (no title, publication date, or excerpt): https://llamaindex.ai/blog/parse-ocria-1

    Skipped (no title, publication date, or excerpt): https://llamaindex.ai/blog/ocr-for-kyc

    Skipped (no title, publication date, or excerpt): https://llamaindex.ai/blog/mortgage-document-automation

    Skipped (no title, publication date, or excerpt): https://llamaindex.ai/blog/Income-verification-api

    Skipped (no title, publication date, or excerpt): https://llamaindex.ai/blog/kyc-automation

    Skipped (no title, publication date, or excerpt): https://llamaindex.ai/blog/deep-extraction

    Skipped (no title, publication date, or excerpt): https://llamaindex.ai/blog/ai-document-classification

    Skipped (no title, publication date, or excerpt): https://llamaindex.ai/blog/ocr-accuracy

    Skipped (no title, publication date, or excerpt): https://llamaindex.ai/blog/unstructured-data-extraction

    Skipped (no title, publication date, or excerpt): https://llamaindex.ai/blog/agentic-document-extraction

    Skipped (no title, publication date, or excerpt): https://llamaindex.ai/blog/agentic-document-processing

    Skipped (no title, publication date, or excerpt): https://llamaindex.ai/blog/ocr-for-tables

    Inspected 12 index links in publisher order

  • Simon Willison
    30 dated entries parsed

    0 entries skipped (missing metadata or outside first-party scope)

At most 150 feed entries, 12 article pages per source, and 2 MB per response are inspected. Articles missing publication dates or excerpts are skipped. Hugging Face community posts are excluded. Date-only publication metadata is interpreted as midnight UTC. Coverage is limited to the configured publishers, not the entire web. Failed sources are not treated as having no news.

Filtering this refresh: {"duplicate": 0, "irrelevant": 41, "outside_window": 236, "recently_picked": 0}