λagent brief
latest

Daily briefing for the agentic web.

6-min daily brief in your inbox every morning. Full analysis free on the web.

.agent >/commands, /sign-up...
issues243
time saved86.8k min
sources434.4k
∷archive
see rest →
  1. FriNov20
    Cover for Cheap Models, Costly Failures

    Cheap Models, Costly Failures

    Cheap Frontier Claims Cognition's evaluation reportedly puts Gemini 4 Neon within one point of Opus 5.5 at roughly a third the cost. · Benchmarks Don't Hold MAST attributes 41.77% of multi-agent failures to specification issues; no frontier model tops 50% on ITBench-AA. · Architecture Is Risk HF's July intrusion post-mortem points to sandbox breakouts and shared credentials, not model capability.

    description

    • Cheap Frontier Claims Cognition's evaluation reportedly puts Gemini 4 Neon within one point of Opus 5.5 at roughly a third the cost.
    • Benchmarks Don't Hold MAST attributes 41.77% of multi-agent failures to specification issues; no frontier model tops 50% on ITBench-AA.
    • Architecture Is Risk HF's July intrusion post-mortem points to sandbox breakouts and shared credentials, not model capability.
    Abundant LabsAmazonAnthropic+47
    425m saved3072 sources24 min read
  2. ThuNov19
    Cover for Agent Swarms Meet Supply Chain Risk

    Agent Swarms Meet Supply Chain Risk

    Coordination Primitive Vercel's "vercel board" gives agent swarms durable, permissioned message passing — a primitive Guillermo Rauch ties to a notable cyber feat. · Supply Chain Alarm Two malicious Microsoft agent-framework PyPI releases shipped a credential stealer for 33 minutes; builders advised to pin 1.26.0. · Failure Anatomy IBM/Berkeley's MAST taxonomy and Meta's Gaia2 (~42% frontier ceiling) shift evals from single scores to why agents fail.

    description

    • Coordination Primitive Vercel's "vercel board" gives agent swarms durable, permissioned message passing — a primitive Guillermo Rauch ties to a notable cyber feat.
    • Supply Chain Alarm Two malicious Microsoft agent-framework PyPI releases shipped a credential stealer for 33 minutes; builders advised to pin 1.26.0.
    • Failure Anatomy IBM/Berkeley's MAST taxonomy and Meta's Gaia2 (~42% frontier ceiling) shift evals from single scores to why agents fail.
    AMDAWSAbundant Labs+66
    858m saved3800 sources26 min read
  3. WedNov18
    Cover for Astra Lands, Coding Flatlines

    Astra Lands, Coding Flatlines

    Benchmarks Get Messy OpenAI's GPT-6.2 Astra and Sol land near-frontier on Artificial Analysis's Intelligence Index, while ARC Prize verifies 99.9% on ARC-AGI-3 under its provider adapter harness — a number that moves ~37 points with the harness. · GUI Agents Scoreboard Hugging Face's ScreenSuite bundles 13 GUI benchmarks; H Company reports Holo4 27B at 61.7% on OSWorld 2.0 versus 81.8% for Opus 5.5. · Stack Hardening MCP goes stateless and human gates get reversibility tiers — practical plumbing for shipping agents.

    description

    • Benchmarks Get Messy OpenAI's GPT-6.2 Astra and Sol land near-frontier on Artificial Analysis's Intelligence Index, while ARC Prize verifies 99.9% on ARC-AGI-3 under its provider adapter harness — a number that moves ~37 points with the harness.
    • GUI Agents Scoreboard Hugging Face's ScreenSuite bundles 13 GUI benchmarks; H Company reports Holo4 27B at 61.7% on OSWorld 2.0 versus 81.8% for Opus 5.5.
    • Stack Hardening MCP goes stateless and human gates get reversibility tiers — practical plumbing for shipping agents.
    ARC PrizeAWSAlice Labs+52
    338m saved560 sources28 min read
  4. TueNov17
    Cover for Agents Attack, Agents Accelerate

    Agents Attack, Agents Accelerate

    Threats Go Live Google's Threat Intelligence Group declared "the era of autonomous attacks has arrived"; California sued OpenAI over the July Hugging Face intrusion. · Numbers Under Scrutiny Holotron-12B claims 80.5% WebVoyager on its own harness — a 0.3-point edge over Holo2-8B. · Tools Move Out MCP's ARD spec pulls tool selection outside the model, with Microsoft, Google, and AWS publishing on it.

    description

    • Threats Go Live Google's Threat Intelligence Group declared "the era of autonomous attacks has arrived"; California sued OpenAI over the July Hugging Face intrusion.
    • Numbers Under Scrutiny Holotron-12B claims 80.5% WebVoyager on its own harness — a 0.3-point edge over Holo2-8B.
    • Tools Move Out MCP's ARD spec pulls tool selection outside the model, with Microsoft, Google, and AWS publishing on it.
    AWSAirbnbAlibaba+33
    353m saved542 sources16 min read
  5. MonNov16
    Cover for Agent Threat Meets Agent Scale

    Agent Threat Meets Agent Scale

    Capability Push DeepSeek-V4 claims a million-token context for agents; the usable retrieval curve, not the window, is the real question. · Security Edge The FBI's ARTEX advisory and Microsoft's Storm-0940 attribution to a ConnectWise compromise remain vendor and threat-intel claims. · Infra Matures OpenEnv's Linux Foundation handoff signals agent tooling consolidating — while the attack surface sharpens.

    description

    • Capability Push DeepSeek-V4 claims a million-token context for agents; the usable retrieval curve, not the window, is the real question.
    • Security Edge The FBI's ARTEX advisory and Microsoft's Storm-0940 attribution to a ConnectWise compromise remain vendor and threat-intel claims.
    • Infra Matures OpenEnv's Linux Foundation handoff signals agent tooling consolidating — while the attack surface sharpens.
    AWSAlibabaAmazon+55
    456m saved583 sources23 min read