Thesis

From record-fetch to work-moving agents

Architecture · Vertical Depth · Execution

General models were the starting gun. Value accrues to architecture: vertical agent systems that reason, decide, and execute inside real workflows.

Problem

Most "agents" are UI over record retrieval. They fetch context then hand work back to humans. Value stalls at the point of execution.

Thesis

  1. Architecture beats features. Durable advantage comes from systems design that combines memory, tools, policies, and verification.
  2. Vertical depth beats horizontal breadth. Industry language, incentives, and compliance create moats that general copilots cannot cross.
  3. Execution closes the loop. Agents must initiate actions, not just summarize. The loop is perceive → decide → act → verify → learn.
  4. Shared infrastructure compounds. Common rails for security, observability, commerce, and data contracts reduce time to market across studios.
  5. Operator networks accelerate truth. Domain operators plus model engineers produce higher signal and lower CAC than isolated R&D.

AgentOS: the reference architecture

Goal: move work forward safely.

1. Interface Layer

  • Channels: chat, API, scheduled jobs, event hooks
  • Identity: user, workspace, role, permissions

2. Reasoning Layer

  • Planning: multi-step decomposition, tool-aware plans
  • Memory: long-term profile, short-term scratchpad, case libraries
  • Policy: constraints, instruction hierarchy, alignment rules

3. Action Layer

  • Tools: CRUD in SaaS, RPA, retrieval, generation, payments
  • Transactions: idempotency keys, rollback, human-in-the-loop gates
  • Verification: invariants, unit checks, reconciliation against source-of-truth

4. Data and Contracts

  • Typed schemas. Versioned prompts. Test fixtures. Synthetic sandboxes.
  • Event bus for audit, telemetry, and replays.

5. Observability

  • Traces of thoughts and tools. Success and safety metrics. Drift alarms.
  • Experiment switches and blue-green prompts.

6. Governance

  • PII minimization. Key vaults. Policy packs per vertical.
  • Evidence logs for SOC 2, HIPAA, FERPA, PCI where relevant.

See the Difference

Watch how architecture changes outcomes

❌ Record-Fetch Agent
UI over retrieval. Work stops at the human.
User: Process the vendor invoice from Acme Corp
0s
Time
0
Actions
Pending
Status
✓ Work-Moving Agent
Full loop: perceive → decide → act → verify → learn
User: Process the vendor invoice from Acme Corp
0s
Time
0
Actions
Pending
Status

The Five Laws of Work-Moving Agents

1. Task specificity increases success probability. Narrow verbs outperform vague intents.

2. Tool determinism enables trust. The agent should know what success looks like before it acts.

3. State awareness prevents loops. Plans must check current system state before execution.

4. Tight feedback compounds learning. Every action produces a labeled outcome that updates policy and memory.

5. Human cadence creates leverage. Insert approvals only where risk or ambiguity is high.

Why Verticalization Wins

  • Language: each industry has stable jargon, forms, and flows that improve retrieval and planning.
  • Incentives: agents tuned to how value is realized in the vertical select better actions.
  • Integrations: a few deep systems beat a thousand shallow OAuths.
  • Regulatory tailwinds: compliance scaffolding is reusable and defensible.

Studio Playbook

  1. Fieldwork first: shadow operators. Map the canonical workflow. Identify three verbs that produce money or time.
  2. Thin wedge: ship a single agent that owns one verb end-to-end.
  3. Close the loop: instrument verification and reconciliation on day one.
  4. Scale by adjacency: add neighbors in the workflow graph once the core loop is reliable.