{
  "_meta": {
    "publication": "Agent Techniques Weekly",
    "schemaVersion": "2026.05.02",
    "generatedAt": "2026-09-20T15:08:42.067Z",
    "canonicalUrl": "https://brianletort.ai/industry/agents/2026-W19",
    "markdownUrl": "https://brianletort.ai/industry/agents/2026-W19/llm.md",
    "sourceFile": "src/data/industry/agents/2026-W19.ts"
  },
  "issue": {
    "slug": "2026-W19",
    "isoYear": 2026,
    "isoWeek": 19,
    "issueNumber": 3,
    "publishedAt": "2026-05-09",
    "cadence": "weekly",
    "periodLabel": "Week 19 of 2026",
    "bigRead": {
      "headline": "Agentic work became a review problem: the scarce skill is defining good and done.",
      "body": "W19's technique read is verification-first delegation. As coding agents, research agents, and office copilots take more steps on their own, the bottleneck moves to checks: tests, source trails, rubric reviews, approval gates, and separate critic passes. The teams getting real leverage are not asking agents to do more blindly; they are making the definition of done machine-checkable wherever possible."
    },
    "technique": {
      "name": "Verifier-first workflows",
      "mode": "build",
      "summary": "Define the acceptance check before giving the agent meaningful autonomy.",
      "whyItMatters": "Agents are now good enough to produce plausible work quickly. Without a verifier, speed compounds errors; with one, it becomes leverage.",
      "anatomy": [
        {
          "label": "Goal",
          "detail": "State the outcome and the boundary of delegated work."
        },
        {
          "label": "Context",
          "detail": "Give the agent the sources, files, examples, and constraints it needs."
        },
        {
          "label": "Tools",
          "detail": "Limit actions to the connectors, commands, and systems required for the job."
        },
        {
          "label": "Verifier",
          "detail": "Define how the output is checked before it is trusted."
        },
        {
          "label": "Escalation",
          "detail": "Name what requires human review, approval, or rollback."
        }
      ],
      "examples": [
        "Unit-test-gated coding agents",
        "source-grounded research briefs",
        "critic subagents"
      ],
      "source": "Agent Harness Design Patterns",
      "sourceUrl": "https://zylos.ai/research/2026-03-31-agent-harness-design-patterns/"
    },
    "agentCapabilities": [
      {
        "vendor": "OpenAI",
        "product": "Codex",
        "mode": "build",
        "date": "2026-05-09",
        "capability": "Workspace-bound coding assistance increasingly organized around issues, tests, review, and automation hooks.",
        "meaning": "The durable read is not any single coding model; it is the emergence of a build harness where context, tools, and verification are part of the workflow.",
        "source": "OpenAI Codex goals documentation",
        "sourceUrl": "https://developers.openai.com/codex/use-cases/follow-goals"
      },
      {
        "vendor": "Anthropic",
        "product": "Claude / Claude Code",
        "mode": "cowork",
        "date": "2026-05-09",
        "capability": "Persistent instructions, skills, and agentic coding workflows point toward reusable professional work patterns.",
        "meaning": "Claude-style workflows are strongest when the human supplies policy, examples, and review criteria that survive beyond one chat.",
        "source": "Agent Harness Engineering and Agentic Loops",
        "sourceUrl": "https://genalphai.com/agentic-loops-and-harness-engineering/"
      },
      {
        "vendor": "Microsoft",
        "product": "Copilot / Scout",
        "mode": "automate",
        "date": "2026-05-09",
        "capability": "The Microsoft agent surface is moving toward M365-grounded action, background coordination, and enterprise controls.",
        "meaning": "The enterprise differentiator is governed access to mail, calendar, files, identity, and policy, not generic answer quality.",
        "source": "OpenAI Codex goals documentation",
        "sourceUrl": "https://developers.openai.com/codex/use-cases/follow-goals"
      }
    ],
    "skillsAndConnectors": [
      {
        "ecosystem": "MCP-capable agents",
        "name": "Connector-backed workflow",
        "type": "connector",
        "date": "2026-05-09",
        "signal": "Agents increasingly use connectors to reach source systems instead of relying on pasted context.",
        "why": "Connectors turn chat into work by letting agents read the system of record and return traceable output.",
        "source": "Agent Harness Engineering and Agentic Loops",
        "sourceUrl": "https://genalphai.com/agentic-loops-and-harness-engineering/"
      },
      {
        "ecosystem": "Claude / Cursor / Codex / Hermes",
        "name": "Reusable skills",
        "type": "skill",
        "date": "2026-05-09",
        "signal": "Repeated workflows are being packaged as skills, rules, plugins, and templates.",
        "why": "Skills are the portability layer for operating knowledge; they prevent teams from re-teaching the same workflow every week.",
        "source": "Agent Harness Engineering and Agentic Loops",
        "sourceUrl": "https://genalphai.com/agentic-loops-and-harness-engineering/"
      }
    ],
    "proofOfValue": [
      {
        "actor": "Practitioner teams",
        "workflow": "Coding, research, operations, and professional drafting",
        "evidence": "benchmark",
        "claim": "The strongest reported wins come from bounded workflows with clear checks, not open-ended autonomy.",
        "read": "Treat value claims as credible when the workflow, baseline, and verifier are visible. Treat broad percentage claims without methods as directional at best.",
        "source": "Agent Harness Design Patterns",
        "sourceUrl": "https://zylos.ai/research/2026-03-31-agent-harness-design-patterns/"
      }
    ],
    "enterpriseReadiness": [
      {
        "dimension": "permissioning",
        "note": "Agents need least-privilege access scoped to the workflow, not broad user-equivalent authority by default."
      },
      {
        "dimension": "verification",
        "note": "Every higher-autonomy workflow needs a deterministic check, source trail, rubric review, or human approval gate."
      },
      {
        "dimension": "auditability",
        "note": "Background agents should produce inspectable logs of prompts, tool calls, approvals, outputs, and state changes."
      },
      {
        "dimension": "cost",
        "note": "Loops need budgets and stop conditions because repeated agent calls can turn productivity experiments into runaway spend."
      }
    ],
    "scorecard": {
      "asOf": "2026-05-09",
      "rows": [
        {
          "mode": "chat",
          "leadingPattern": "Structured context and critique loops",
          "representativeTools": [
            "ChatGPT",
            "Claude",
            "Copilot Chat"
          ],
          "controlGap": "Quality still depends on the user's review discipline."
        },
        {
          "mode": "cowork",
          "leadingPattern": "Human-supervised delegation with persistent project memory",
          "representativeTools": [
            "Claude Cowork",
            "Microsoft Copilot",
            "Cursor"
          ],
          "controlGap": "State, approvals, and source grounding must be visible."
        },
        {
          "mode": "build",
          "leadingPattern": "Workspace-bound agents with tools, tests, and worktrees",
          "representativeTools": [
            "Claude Code",
            "Codex",
            "Cursor",
            "OpenCode"
          ],
          "controlGap": "Verification quality determines whether speed becomes rework."
        },
        {
          "mode": "automate",
          "leadingPattern": "Scheduled loops with state, tools, and escalation gates",
          "representativeTools": [
            "Codex Automations",
            "Microsoft Scout",
            "Hermes",
            "OpenClaw"
          ],
          "controlGap": "Always-on agents need identity, audit trails, budgets, and stop conditions."
        }
      ]
    },
    "tryThis": {
      "title": "Run a Verifier-first workflows experiment",
      "steps": [
        "Pick one recurring weekly task with a clear definition of done.",
        "Write a one-page loop contract: goal, context, tools, verifier, stop condition, and escalation rule.",
        "Run it manually once with an agent and record where the verifier was weak.",
        "Only automate the task after the verifier catches the most likely failure mode."
      ],
      "expectedOutcome": "A reusable workflow contract and a clearer read on whether the task is ready for cowork, build, or automate mode."
    },
    "watchlist": [
      {
        "window": "Next 7 days",
        "title": "Copilot and Scout agent releases",
        "why": "Microsoft's advantage is governed enterprise context; any new background or M365 action capability changes the automation surface."
      },
      {
        "window": "Next 7 days",
        "title": "Claude and Codex skill ecosystems",
        "why": "Reusable skills and plugins are the leading indicator that agentic workflows are becoming products, not prompts."
      },
      {
        "window": "Next 30 days",
        "title": "Hermes, OpenClaw, OpenCode, and adjacent OSS harnesses",
        "why": "Open-source harnesses reveal which control points matter most: memory, channels, terminal build loops, or automations."
      },
      {
        "window": "Next 30 days",
        "title": "Evidence-backed value claims",
        "why": "The newsletter should elevate wins with named workflows, baselines, and verification methods, not generic productivity claims."
      }
    ],
    "changelog": [
      "Backfilled Agent Techniques Weekly issue 03 for Week 19 of 2026."
    ]
  }
}