{
  "schemaVersion": "shareplane.claim-evidence-register.v1",
  "workId": "artifact:the-bottleneck-moved",
  "title": "The Bottleneck Moved",
  "evidenceCutoff": "2026-10-03",
  "posture": "Experience-backed engineering thesis. External evidence supports mechanisms and directional claims; counterevidence is retained where it limits generalization.",
  "claims": [
    {
      "id": "claim:implementation-cheaper",
      "text": "AI is reducing the marginal effort required to produce many classes of software implementation.",
      "posture": "strong industry observation",
      "support": [
        "source:dora-2026-tensions",
        "source:openai-agents-api-2026"
      ],
      "boundary": "Magnitude varies by task, developer, repository maturity, toolchain, and model generation."
    },
    {
      "id": "claim:system-amplifier",
      "text": "The surrounding engineering and organizational system materially shapes the value produced by AI-assisted development.",
      "posture": "externally supported",
      "support": [
        "source:dora-2025",
        "source:anthropic-tools",
        "source:openai-agents-api-2026"
      ],
      "boundary": "No single source establishes a universal optimal organizational design."
    },
    {
      "id": "claim:rigor-migrates-upward",
      "text": "As generation gets cheaper, more engineering effort moves toward validation, context, system design, authority, and verification.",
      "posture": "thesis supported by converging evidence",
      "support": [
        "source:dora-2026-tensions",
        "source:nist-ai-rmf"
      ],
      "boundary": "The degree of migration differs across safety, regulatory, performance, and operational contexts."
    },
    {
      "id": "claim:productivity-contextual",
      "text": "AI coding productivity is not a single universal speedup number.",
      "posture": "strongly supported",
      "support": [
        "source:metr-2025-rct",
        "source:metr-2026-update",
        "source:metr-2026-task-substitution"
      ],
      "boundary": "Results change with models, tasks, workflow, user adaptation, and task substitution."
    },
    {
      "id": "claim:judgment-scarcity",
      "text": "Judgment increasingly becomes the limiting factor as implementation becomes more abundant.",
      "posture": "Tony Malott thesis / forward inference",
      "support": [
        "source:dora-2025",
        "source:dora-2026-tensions"
      ],
      "boundary": "Not established as a universal labor-market fact."
    }
  ],
  "sources": [
    {
      "id": "source:dora-2025",
      "kind": "Independent industry research",
      "title": "DORA 2025: State of AI-assisted Software Development",
      "url": "https://dora.dev/research/2025/dora-report/",
      "summary": "DORA characterizes AI as an amplifier of an organization’s existing strengths and weaknesses, with the largest returns coming from the underlying organizational system rather than the tool alone.",
      "caveat": "Survey-based organizational research does not prove the article’s broader forecast about team size or role displacement."
    },
    {
      "id": "source:dora-2026-tensions",
      "kind": "Independent industry research",
      "title": "DORA: Balancing AI tensions",
      "url": "https://dora.dev/insights/balancing-ai-tensions/",
      "summary": "DORA reports that AI accelerates initial code generation while time savings are often reallocated to auditing and verification; higher AI adoption is associated with greater throughput and greater instability.",
      "caveat": "Observed tradeoffs across organizations are not a universal causal law."
    },
    {
      "id": "source:metr-2025-rct",
      "kind": "Counterevidence / randomized trial",
      "title": "METR: Early-2025 AI and experienced open-source developers",
      "url": "https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/",
      "summary": "In a randomized trial of 16 experienced developers completing 246 tasks in mature repositories, early-2025 AI tools increased completion time by 19 percent.",
      "caveat": "METR warns against generalizing the result to most development, and the tested tools reflect early-2025 capabilities."
    },
    {
      "id": "source:metr-2026-update",
      "kind": "Capability-change evidence",
      "title": "METR: Developer productivity experiment design update",
      "url": "https://metr.org/blog/2026-02-24-uplift-update/",
      "summary": "METR says it is likely developers were more accelerated by AI in early 2026 than in early 2025, but calls the magnitude weakly evidenced because of selection effects.",
      "caveat": "The follow-up does not establish a reliable universal speedup estimate."
    },
    {
      "id": "source:metr-2026-task-substitution",
      "kind": "Measurement nuance",
      "title": "METR: Task Substitution and Uplift",
      "url": "https://metr.org/blog/2026-05-08-task-substitution-and-uplift/",
      "summary": "METR argues that AI productivity depends partly on which tasks people choose once AI is available, so speed on old tasks and value from new tasks are not the same measure.",
      "caveat": "This is a measurement framework, not direct proof of the article’s role-compression thesis."
    },
    {
      "id": "source:anthropic-context",
      "kind": "Primary technical guidance",
      "title": "Anthropic: Effective context engineering for AI agents",
      "url": "https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents",
      "summary": "Anthropic frames context as a critical finite resource and context engineering as curating the high-signal state, tools, and information required for effective long-horizon agents.",
      "caveat": "Vendor engineering guidance is experience-based and not an independent productivity study."
    },
    {
      "id": "source:anthropic-tools",
      "kind": "Primary technical guidance",
      "title": "Anthropic: Writing effective tools for agents",
      "url": "https://www.anthropic.com/engineering/writing-tools-for-agents",
      "summary": "Anthropic argues that agent effectiveness depends strongly on tool design, meaningful returned context, evaluations, and clear functionality boundaries.",
      "caveat": "Supports system architecture claims, not the employment implications of the thesis."
    },
    {
      "id": "source:openai-agents-api-2026",
      "kind": "Primary platform evidence",
      "title": "OpenAI: Introducing the Agents API",
      "url": "https://openai.com/index/introducing-the-agents-api/",
      "summary": "OpenAI says useful long-running agents need a harness that manages context, tools, and subagents, plus persistent environments where agents can work with files, run code, and save intermediate results.",
      "caveat": "Provider evidence does not independently validate productivity or labor-market predictions."
    },
    {
      "id": "source:nist-ai-rmf",
      "kind": "Standards / risk framework",
      "title": "NIST AI RMF Core: Measure and Manage",
      "url": "https://airc.nist.gov/airmf-resources/airmf/5-sec-core/",
      "summary": "NIST calls for testing before deployment and during operation, rigorous performance assessment, documented measurement, independent review, and risk-response controls.",
      "caveat": "NIST provides a voluntary risk framework, not a prescriptive architecture for autonomous software engineering."
    }
  ]
}