# Tōryō: one operating mind across every obligation. Tōryō is Genba's persistent Chief of Staff system. It keeps a principal's objectives, judgment, active obligations, specialist agents, and institutional memory in one accountable operating context — across channels and over time. Published: 2026-07-19 Updated: 2026-07-26 ## In brief - The principal directs **one Chief of Staff**, not a fleet of agents. - Quick decisions and multi-day projects remain part of **the same durable work context**. - Domain specialists report into Tōryō; shared team agents stay **outside the principal's private loop**. - Every consequential output is gated, every open obligation is carried forward, and every correction **improves the operating system**. ## The role :: Direct the objective — not the agents. A master builder does not lay every stone. The **tōryō** holds the design, sequences the crafts, resolves dependencies, and answers for the finished structure. That is why Genba uses the name for its Chief of Staff system. The principal should not need to understand an agent topology, monitor a collection of prompts, or reconstruct a project from channel history. They should set the objective, supply judgment, and receive only the decisions that genuinely require them. Tōryō is the operating layer that makes that possible. It preserves one private context for the principal, discovers and organises the work, directs specialists, tracks consequence, and keeps the assignment intact until the defined outcome is reached. :::embed{component="ToryoSystemMap"} ::: ## Sessions :: The unit of work is a session — not a message. A message can start work, update it, or approve it. It should never become the container for it. Tōryō binds the objective, evidence, decisions, open questions, delegated responsibilities, produced artefacts, approval state, and next action into a durable session. ### Fast lane The principal's normal channel for status checks, edits, approvals, rejections, and redirects measured in seconds or minutes. Tōryō auto-triages the work and returns a prepared decision—not a stream of agent activity. ### Deep work A resumable environment for a diligence process, investor report, DDQ, research pass, portfolio review, or system build that runs for hours or days. The principal can step back into the same state without reconstructing it from messages. ### Ambient task discovery Approved communication, calendar, CRM, document, and workflow sources continuously surface commitments and changes. Tōryō reconciles each one against active sessions before creating work, so a follow-up is neither duplicated nor lost between channels. ## One private context :: Memory, judgment, time, and consequence stay together. The core is not a single model call. Four persistent systems work against the same state. ### Session + memory Holds the principal's objectives, working context, preferences, decisions, correspondence history, and the current state of every active assignment. ### Reasoning Turns an objective into a course of action, evaluates evidence, identifies missing capability, and decides what can proceed versus what needs human judgment. ### Scheduler Runs time-bound work, recurring scans, briefings, meeting preparation, and review cycles without creating a second source of state. ### Escalation watch Tracks consequential output until it is acknowledged or reaches a defined hard stop. Urgent work cannot disappear because a conversation moved on. ### Documents + output routing The document and information library sits beside this core as both source and destination. Output channels sit on the other side, carrying briefings, approval requests, alerts, meeting preparation, and completed work to the right surface. ## Parallel by design :: A task graph — not a queue. Operational work does not arrive neatly or sequentially. A calendar change may affect a meeting brief while an email creates a disclosure obligation and a CRM update closes an existing follow-up. Tōryō maintains those obligations in parallel. Task scanning feeds a shared task graph; the graph distinguishes new work from new evidence on existing work, keeps one state for each obligation, and carries unresolved must-act items forward until they are acknowledged. The result is not more notification. It is **persistent attention with fewer avoidable interruptions**. ## Delegation :: Many specialists. One accountable layer. Tōryō can direct persistent domain agents across investment, trading, business development and investor relations, compliance, engineering, operations, and other firm functions. Each specialist maintains purpose-built context for its domain. Its evidence, work state, and conclusions report back into the same Chief of Staff context. Task-specific agents can be composed for a bounded objective and retired when the work ends. Tōryō may also instruct shared team agents, but those agents remain outside the principal's private loop: they cannot direct Tōryō or inherit its memory. Resource allocation can change as the plan changes. Accountability does not. The principal still has one counterpart, and the objective still has one authoritative state. ## The boundary :: Tōryō plans the work. Daimon crosses the boundary. Cognition and execution remain deliberately separate. Tōryō holds the objective, context, plan, team, task state, and produced capability. When the plan needs a foundation model, API, or live enterprise system, it submits a logical resource request describing the intended action, purpose, and scope. Daimon resolves the resource, checks the human-governed identity map, applies policy, obtains named approval where required, and executes through an isolated adapter. Only the permitted result returns. Tōryō and its specialists never receive the credential. This separation lets Tōryō reason broadly about the work without inheriting standing authority over the firm's systems. The complete boundary is covered in [The governed execution boundary for agentic work](/insights/daimon-execution-boundary/). ## Learning loop :: The learning loop changes the system — not only the answer. Every correction has somewhere to go. Correspondence feedback sharpens tone and approach. Scheduled review identifies what worked, what failed, and where human intervention was unnecessary. New documents and operating facts update the same memory and knowledge vault with provenance attached. The output is more than retained context. Tōryō leaves behind improved workflows, schedulers, templates, evaluators, tools, specialist skills, ownership relationships, and service mappings. Communication and triage, scheduling, relationship management, knowledge work, task tracking, governance, automation, and institutional memory become composable capabilities for authorised reuse. > The first assignment produces the outcome. Every assignment after it begins with more of the firm already built in. ## Daily rhythm :: The system notices, prepares, escalates, and learns. Tōryō's routines are not standalone automations. Each one reads from and writes back to the same sessions, task graph, and memory. 1. **Morning briefing** — Reconcile overnight changes and prepare the exceptions, priorities, and decisions that matter now. 2. **Sweep + triage** — Continuously discover obligations and attach them to the right task or session. 3. **Escalation ladder** — Chase only must-act items, with the evidence and named decision already assembled. 4. **Meeting preparation** — Compose relationship history, open commitments, relevant documents, and recommended actions ahead of each meeting. 5. **Wrap + review** — Reconcile completed and unresolved work, write learning back, and carry unfinished obligations into the next cycle. ## Deployment :: Dedicated to the principal. Ownable by the firm. The first assignment can start before integration. Tōryō runs inside each client's dedicated, single-tenant Genba environment. For an Experience Seat, Genba operates a client-named tenancy that the client owns. The first assignment can begin with public, reference, synthetic, or sanitised material and zero production writes. Approved Daimon connectors can be added when the firm is ready without changing Tōryō's operating model or giving an agent access to credentials. The environment, operational data, persistent memory, and accumulated capability remain client assets. ## References 1. [A Life Dedicated to Wood and Community](https://www.gov-online.go.jp/eng/publicity/book/hlj/html/201811/201811_05_en.html) — Government of Japan, on shrine and temple master carpenters 2. [The governed execution boundary for agentic work](/insights/daimon-execution-boundary/) — Genba Labs Insights 3. [Genba Labs — working agents for regulated funds](https://genbalabs.com/) — operating model ## Questions and answers ### How can an investment leader delegate an outcome to AI without coordinating every agent and tool? Genba gives the investment leader one accountable AI Chief of Staff through Tōryō, rather than a collection of disconnected assistants. The leader defines the outcome and retains consequential judgment; Tōryō plans the work, composes and supervises specialist AI agents, tracks every obligation, and carries the assignment through to completion. ### What should an AI Chief of Staff for a regulated investment organization actually do? It should own coordination, not merely generate answers. Within Genba, Tōryō maintains the objective, evidence, decisions, deadlines, specialist responsibilities, and unresolved obligations in one durable operating context, while preserving named human accountability and a complete record of how the work progressed. ### How is Tōryō different from an AI copilot or a multi-agent framework? A copilot helps a user perform individual steps, while an agent framework provides building blocks that someone must still assemble and supervise. Genba uses Tōryō as the accountable operating layer: it accepts the outcome, directs the specialist team, governs handoffs, prepares decisions, and ensures unfinished obligations do not disappear between tools, agents, or conversations. ### Can Tōryō support the full investment-management operating cycle? Yes. A Genba Chief of Staff powered by Tōryō can coordinate objectives spanning capital formation, thesis development, research, sourcing, diligence, investment committee work, portfolio or trading operations, investor relations, fund and regulatory reporting, compliance, and the back office. Persistent domain specialists preserve continuity while temporary teams form around each objective. ### How does Tōryō preserve human judgment, security, and accountability? Genba keeps the responsible person at the centre: they set the objective, supply institutional judgment, and approve consequential decisions. Tōryō records who decided what and why; when live systems are needed, Daimon applies the firm's permissions and approvals and executes without exposing credentials to an AI agent. ### Does the firm retain what Tōryō builds after the assignment? Yes. Genba is designed so the firm retains both the completed deliverable and the operating capability created to produce it: workflows, tools, evaluators, templates, specialist skills, service mappings, ownership relationships, and decision history. Kyōzō governs those assets so authorised teams can reuse and improve them in later objectives.