Control plane for agent organizations

Pilot

Operate AI agents as an organization with responsibilities and budgets—not a collection of tools.

From company objectives to teams, roles, work, budgets, approvals, and artifacts, Pilot shows leaders why agents are working, what resources they are using, and where a human decision is required.

Discuss an AI operating model
A Pilot operating board joining company objectives, agent responsibility, budget, approval, and artifacts
  1. A person defines the company objective, time horizon, and success conditions.
  2. Agents propose strategy, teams, roles, and a resource plan.
  3. A responsible leader confirms autonomy limits, approval gates, and budget ceilings.
  4. Connected runtimes perform work and report progress, cost, and results.
  5. Leaders review artifacts, exceptions, and pending approvals, intervening where required.
The first five minutes of a company

Define an objective and receive a strategy and organization proposal that can be reviewed.

“Build an AI quality-control company for Korean manufacturers and complete the first field validation within 90 days.”
  1. 01
    Create the company and top-level objective

    The company becomes the operating context to which every piece of work can connect.

  2. 02
    A CEO agent proposes strategy

    Market hypothesis, product scope, 90-day priorities, teams, and budget are presented at a board-review level.

  3. 03
    A person approves strategy and boundaries

    People confirm objectives, budget ceilings, permitted autonomy, and decisions that always require approval.

  4. 04
    Form teams and assign roles

    Agents receive explicit responsibilities and connect to the runtimes that will perform the work.

  5. 05
    Operate work and artifacts

    Review current work, spend, pending approval, failure, and completed artifacts by objective and owner.

After the first few agents

More chat windows do not create an organization.

Running one agent differs from aligning many agents to a business objective. If leaders cannot see ownership, strategic relevance, cost, or pending human decisions, autonomy becomes operating uncertainty.

Pilot is not a cosmetic employee metaphor. It creates the links between company objective, strategy, team, work, budget, approval, and artifact. Leaders inspect business state and exceptions rather than reading every prompt.

Questions a leader must answer at a glance

See who is doing what, why it matters, what it costs, and what awaits a decision.

01

Who owns the work?

Inspect agent role, team, current assignment, and responsibility through one organizational model across runtimes.

02

Why does it matter?

Trace work back through team objectives to company strategy and priority.

03

Where are time and money going?

Expose budget and actual usage by team, objective, and task to avoid hidden or unbounded consumption.

04

What needs a human decision?

Separate strategy changes, budget increases, external actions, and high-risk execution from routine progress.

From objective to artifact

Every agent’s work remains connected to the direction the company chose.

Producing more tasks is not performance. Pilot connects the origin, owner, resource boundary, and final artifact.

  1. 01

    Company objective

    Define the outcome, time horizon, success condition, and human-controlled boundary.

    company · objective · constraints
  2. 02

    Strategy and team objective

    Agents propose strategy and subgoals; the board or owner approves priorities and resources.

    strategy · team goal · board decision
  3. 03

    Accountable work

    Each task carries an objective link, owner, budget, state, and required approvals.

    owner · task · budget · approval
  4. 04

    Runtime execution

    Connected runtimes execute and report progress, usage, errors, and results.

    runtime · heartbeat · usage · result
  5. 05

    Artifact and intervention

    Review the deliverable first, then handle drift, failure, overspend, and approval waits.

    artifact · exception · intervention
Principles of the operating view

Keep execution detail available, but make outcomes and intervention points the default.

01

Board-level abstraction

Understand progress, resources, approvals, and exceptions without reading every prompt or tool call.

02

Artifacts first

Lead with what was produced and its review state rather than how long an agent appeared to work.

03

Progressive detail

Descend into teams, tasks, sessions, and raw records only when the decision requires it.

04

Bounded autonomy

Budgets, authority, approval gates, stop conditions, and intervention let agents move quickly within an explicit scope.

Current product status

Pilot is pre-release. Product direction and present scope are stated separately.

The operating model is defined, but we do not represent Pilot as a completed self-service product. Current work prioritizes organizations willing to validate real operating scenarios.

Product stage
Pre-release

No promise of public sign-up or immediate production readiness.

Core objects
Company, objective, agent, team, work, budget, approval, and artifact
Operating invariant
Work traces to company objectives; spend and approval waits remain visible
Runtime relationship
A control plane coordinating multiple agent runtimes

Pilot does not replace the runtime that executes the work.

Adoption mode
Design-partner review around a representative AI organization scenario
Separate runtime from organization

Change execution technology without losing company objectives and operating history.

Runtimes execute tasks. Pilot records why the execution exists and how responsibility, budget, and approval apply.

Single team

Validate roles, work, approvals, and spend around one objective and a small agent group.

  • Explicit roles and objective
  • Work and budget tracking
  • Human approval queue

Multi-team company

Connect team objectives, budgets, and runtimes through one company operating model.

  • Organization and responsibility
  • Resources by team and objective
  • Cross-team dependency and exception

Runtime-agnostic control

Translate status and outcomes from different runtimes into a common operating language.

  • Runtime adapters
  • Common work and artifact model
  • Stop, intervene, and recover boundaries
What Pilot is not

Clear boundaries make an adoption decision easier.

Pilot is not a familiar tool repackaged under an agent metaphor.

Not a chatbot

Conversation is only one interface. The core is the operating model for objectives, responsibility, resources, approval, and artifacts.

Not a Jira or GitHub replacement

Pilot connects useful execution surfaces while preserving the higher organizational context for an AI workforce.

Not an agent runtime

Connected runtimes call models and tools. Pilot starts and observes them while applying objectives, budget, and approval boundaries.

Autonomy is not unlimited authority

Agents may plan and act, but only within organizational budgets, permissions, approvals, and stop conditions.

Begin with the objective and responsibility structure of the AI organization you want to operate.

We will review Pilot’s pre-release operating model around a first company objective, agent roles, budget, and approval boundaries.

Ask about the Pilot design-partner program