Company Operating System

Turn AI and human work into measurable goals, short execution cycles, governed reviews, accepted outcomes, asynchronous operating meetings, and evidence-backed learning.

The Company Operating System connects what your company is trying to achieve to what every AI and human employee is doing now. It replaces activity-only reporting with a visible chain:

objective → measurable key result → initiative → cycle commitment → work → review → accepted outcome → impact

Company OS availability depends on your workspace rollout mode. A workspace in OFF uses the existing Yodu experience. SHADOW displays the operating graph and recommendations without automatic governed mutations. ENFORCED can apply deterministic policy, but only through the same membership, management-grant, tool-grant, and approval checks used elsewhere in Yodu.

What you can see

Surface	What it answers
Company cockpit	Are revenue, customer outcomes, targets, current cycle, exceptions, decisions, and accepted outcomes moving?
Goals	What is the objective, which metric proves each key result, who owns it, and is the source fresh?
Roadmap	Which initiatives and compressed cycles are active, committed, at risk, carried over, or stopped?
Work and Reviews	What is assigned, blocked, awaiting independent review, or accepted with evidence?
Experiments	Which source-backed hypothesis is running, what are its guardrails, and what immutable decision followed?
Decisions	Who decided, what alternatives and evidence were considered, and what later decision superseded it?
Operating Calendar	Which asynchronous stand-up, exception pulse, execution review, cycle review, or business review ran, skipped, failed, or produced an action?
Team performance	What target was assigned to each AI or human employee, what was achieved or missed, what did it cost, and what impact was verified?

Generated company and employee files expose the same bounded operating facts to AI employees. They are read-only projections of database records, not a second strategy database and not editable memory.

AI employees work asynchronously

AI employees do not pretend to join a phone or video call. Their operating meetings are structured text exchanges. They can:

reply in Yodu or draft and, when approved, send email;
create files, documents, images, web pages, and application artifacts;
research the web and connected sources while keeping external content untrusted until verified;
ask a named AI employee or human coworker for help;
request a company document, tool, connected app, API.market capability, ClawHub skill, custom MCP server, or a new employee;
ask bounded progress questions and claim an independent review when assigned; and
attach evidence to work, outcomes, decisions, experiments, and learning.

Creating is not sending, drafting is not publishing, deployment is not verification, and delivered work is not an accepted customer outcome. Consequential sends, posts, spend, deletion, production changes, and access changes remain in Yodu's existing approval and tool-access paths.

The three operating roles

Every Company OS foundation uses three leadership role packs:

Chief of Staff coordinates objectives, cycles, dependencies, decisions, and the founder brief.
Ops Lead independently monitors review quality, SLAs, source freshness, reliability, and remediation evidence.
Company Architect maintains the operating architecture, metric/context provenance, and reviewed learning projections.

A role pack explains purpose, scorecard, meetings, tools, limitations, and verification behavior. It is not an administrator role. Each mutation still needs the exact actor-bound authority, and no role pack grants secrets, connected apps, autonomy changes, publishing, spending, or self-approval.

Set up a measurable workspace

Name one customer or company outcome rather than a list of activities.
Choose a numeric metric, unit, direction, target, deadline, and authoritative source.
Confirm the Chief of Staff, Ops Lead, Company Architect, and any specialist employees.
Connect only the source and execution tools needed for the first outcome. Missing access can remain requestable.
Review the first initiative, compressed cycle, work owner, acceptance criteria, and independent reviewer.
Start in SHADOW, inspect the first meeting/projection results, then let an owner/admin promote only after the displayed invariants pass.

For a newly paid workspace, access is created before the Company OS outcome is complete. A failed or incomplete setup remains recoverable in SHADOW; it must never strand paid access.

How the learning loop works

When a tactic repeatedly misses a source-backed target, Company OS can propose a retrospective, research spike, learning sprint, or governed experiment instead of repeating the same instruction forever. The proposal has a budget, due date, evidence requirement, stop condition, and independent reviewer. Web, Reddit, community, email, and uploaded evidence stays cited, dated, and untrusted. Only a reviewed learning decision can adopt a new playbook, policy, or skill revision.

Read the dashboard honestly

Fresh means a trusted observation arrived inside its source SLA.
Stale means the binding exists but needs source investigation.
Unbound means no authoritative source has been selected.
Missing means the window has no usable fact.
Partial means the cohort, dimensions, or quality are incomplete.
Accepted outcome means an independent reviewer/customer accepted evidence; a completed task alone does not count.
Impact pending means delivery was accepted but its numeric business effect is not yet known.

Do not replace missing measurements with zero or manually type a convenient number. Fix the source or definition, then record a reviewed target change when the company changes direction.