Lambda Labs discusses agent infrastructure value and security at Agentic AI Summit, UC Berkeley.

{
  "signal_id": "a7ed327a-649c-4562-8766-1c1bd425c983",
  "signal_type": "linkedin-post-company",
  "signal_subtype": "linkedinPost",
  "detected_at": "2026-08-13T12:40:59.395+00:00",
  "company": {
    "name": "Lambda",
    "domain": "lambdalabs.com"
  },
  "data": {
    "tags": [
      "Artificial Intelligence",
      "Research & Development",
      "Security",
      "Automation"
    ],
    "summary": "Lambda Labs discusses agent infrastructure value and security at Agentic AI Summit, UC Berkeley.",
    "post_url": "https://www.linkedin.com/posts/lambda-cloud_if-we-had-to-compress-lambdas-weekend-at-activity-7490750719675813888-IbUM",
    "num_likes": 51,
    "post_text": "If we had to compress Lambda's weekend at the Agentic AI Summit at University of California, Berkeley into five words: harness is all you need.\n\nThe argument we made on the main stage: agent workloads are machine traffic merging onto roads built for human drivers. The unit of the business is shifting from tokens per dollar to progress per dollar. Swap the model and your investment in it resets. The harness, the verifiers, the traces carry over. That's where the value compounds.\n\nThat's where the moat in agent infrastructure sits: make the unique demands of this workload (sandboxing, systems of record, efficiency across heterogeneous jobs) the default, and integrate them into the stacks people already run. Focused. Specific.\n\nWorkshop with Zachary Mueller: https://lnkd.in/g_HNnSXj\n\nTalk with Chuan Li: https://lnkd.in/gx9Cdxzz\n\nPanel with Chuan Li: https://lnkd.in/gwN9ZZWk\n\n𝗧𝗵𝗿𝗲𝗲 𝗽𝗿𝗼𝗼𝗳𝘀 𝘀𝘂𝗽𝗽𝗼𝗿𝘁𝗶𝗻𝗴 𝗼𝘂𝗿 𝗮𝗿𝗴𝘂𝗺𝗲𝗻𝘁:\n\n𝗦𝗲𝗰𝘂𝗿𝗶𝘁𝘆.\nOur Security competition with Berkeley RDI ran 100,000 battles as a controlled experiment: same model (gpt-oss-20b), escape-proof sandboxes, fair compute for every agent. Everything pinned, and the harness is what won. The attack sharpened round by round, because the harness reads the full battle history and refines each attempt. The defense told the same story in slow motion: a top defender's harness grew from 239 lines...",
    "initiatives": [
      {
        "topic": "investing in agent infrastructure where value compounds",
        "urgency": 0.7
      },
      {
        "topic": "making agent infrastructure demands default and integrated into existing stacks",
        "urgency": 0.8
      },
      {
        "topic": "logging every agent run in The Lab API, now open source",
        "urgency": 0.75
      }
    ],
    "pain_points": [
      {
        "topic": "agent workloads are machine traffic merging onto roads built for human drivers",
        "intensity": 0.5
      },
      {
        "topic": "agents cheat, drift, and burn money",
        "intensity": 0.6
      }
    ],
    "posted_date": "2026-08-05T12:49:02.989Z",
    "num_comments": 0,
    "competitors_mentioned": [
      {
        "name": "Berkeley RDI"
      }
    ],
    "technologies_mentioned": [
      {
        "name": "Lambda harness",
        "status": "using"
      },
      {
        "name": "gpt-oss-20b",
        "status": "using"
      },
      {
        "name": "Kimi K3",
        "status": "using"
      },
      {
        "name": "Nous Research Hermes Agent",
        "status": "using"
      },
      {
        "name": "Claude Code",
        "status": "using"
      },
      {
        "name": "Gemma 4",
        "status": "using"
      },
      {
        "name": "The Lab API",
        "status": "using"
      }
    ]
  }
}

curl https://signals.autobound.ai/v1/signals/a7ed327a-649c-4562-8766-1c1bd425c983 \
  -H "X-API-KEY: $AUTOBOUND_API_KEY"

curl -X POST https://signals.autobound.ai/v1/companies/enrich \
  -H "X-API-KEY: $AUTOBOUND_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"domain":"lambdalabs.com","limit":20}'