Customer-owned AI work systems

Turn a request into finished, reviewable work.

Foundry 41 installs and operates an AI work system in your cloud account. A request in messaging starts an isolated job with the company context and tools you approve, then returns files, decisions, or code for a person to review. You keep the system.

One workflow first30-business-day targetOperate or hand over
ILLUSTRATIVE PRODUCT VIEW · FICTIONAL DATA · BASED ON OUR INTERNAL SKY INTERACTION PATTERN
# leadershipTEAM WORKSPACE · APPROVED MEMBERS
MO

Maya · Operations 9:14 AM

Prepare tomorrow's leadership brief. Compare the pipeline report with last week's decisions, then flag anything without an owner.

41

Foundry agent 9:14 AM

Starting an isolated job with the approved brief template, decision log, and read-only pipeline report.

LEADERSHIP BRIEF V5DECISIONS LOGPIPELINE · READ ONLY
41

Foundry agent 9:18 AM

Tomorrow's brief is ready for review.

LEADERSHIP BRIEF · 08 SEPREADY FOR REVIEW
3RISKS FLAGGED
2DECISIONS NEEDED
4OWNER GAPS
2FILES ATTACHED
REVIEW REQUIRED

Four task-owner updates are proposed. Nothing has been changed yet.

The coordination problem

AI should not create another inbox.

Giving everyone a chatbot can speed up isolated tasks while scattering context, decisions, and outputs across private sessions. Foundry gives the company one governed way to delegate and review real work.

BEFORE · EVERYONE STARTS OVERAFTER · WORK MOVES TOGETHER
Disconnected personal AI compared with coordinated company work On the left, three people use separate chats, files, and business systems. On the right, people and an agent share one work record connected to approved systems and review. SESSION-BY-SESSION AI CUSTOMER-OWNED WORK OPERATIONS FINANCE OWNER PRIVATE CHAT DOC COPY SYSTEM CONTEXT SPLINTERS · DECISIONS DISAPPEAR TEAMMATE REVIEWER SHARED WORKREVIEW READY SYSTEMS ONE OWNER · SHARED CONTEXT · VISIBLE OUTCOME

Before: everyone starts over

Operations, finance, and owners work from separate chats, document copies, and business systems. Context splinters and decisions disappear.

After: work moves together

One request, shared approved context, a named reviewer, and a visible outcome connected to the systems the company chooses.

What we install

Not a chatbot. An operating system for one workflow at a time.

We currently call the installed layer Foundation. It combines company context, bounded tools, isolated execution, human review, and the Software Factory that lets the system improve safely.

01 · COMPANY BRAIN

Give the work the right context—not all the context.

The Company Brain keeps skills, memory, configuration, and approved sources in a form your team can inspect. Each job receives only the relevant context and access.

  • Editable operating knowledge
  • Per-team memory and rules
  • Approved connections to business systems
CONTEXT SELECTIONCLIENT ACCOUNT
How the Company Brain selects context for a jobPolicies, team memory, system data, and an operating skill feed a selected context layer. The bounded context enters an isolated job, which produces a reviewable output. COMPANY SOURCESSELECTED CONTEXTBOUNDED WORK PoliciesAPPROVED FILES Team memoryCHANNEL-SCOPED System dataREAD ACCESS Workflow skillVERSIONED RULES JOB CONTEXTOnly what thistask needsACCESS CHECKED ISOLATED JOB OUTPUTReady to review NOT AN ALL-KNOWING BRAIN · AN INSPECTABLE CONTEXT LAYER

Company sources

Approved policies, team memory, system data, and a versioned workflow skill.

Selected job context

The task receives only the relevant context and access—not the whole company brain.

Bounded work

An isolated job completes the task and returns an output for a person to review.

02 · ISOLATED EXECUTION

Let agents work without giving them the keys to everything.

Each conversation gets a dedicated job environment. The agent receives bounded capabilities, performs the work, and reports back with an artifact or decision—not an invisible chain of autonomous actions.

  • Scoped tools and credentials
  • Separate execution environments
  • Logs and outputs the company keeps
JOB ACTIVITYJOB 041-0927

03 · HUMAN CONTROL

Put people where judgment matters.

Rules decide what the agent may prepare. Named owners handle exceptions, approve consequential actions, and keep the reasoning with the work record.

  • Explicit approval points
  • Named workflow owners
  • Review, release, monitor, and roll back

A concrete first workflow

Start with work people already do every day.

This illustrative clinic example shows the end result: not “chat with AI,” but a visible case that moves from intake to preparation, human judgment, and system outcome.

REFERRAL FOLLOW-UP

One owned case from request to outcome.

The agent prepares routine work. Staff see exceptions and approve action. The record keeps the context, draft, decision, and outcome together.

12FILES REVIEWED
8DRAFTS READY
3MISSING ITEMS
1STAFF DECISION
WORK RECORD · 041-0927OWNER · REFERRAL TEAM
01
Intake opens the case

Approved referral fields and source records enter one work record.

COMPLETE
02
Agent prepares follow-up

Packet checked, missing information flagged, drafts saved.

COMPLETE
03
Staff handles the exception

A named owner reviews the one item outside the approved rule.

REVIEW REQUIRED
04
Approved outcome writes back

Only after approval; the work record retains the decision and output.

PENDING

Why customer-owned matters

Own the operating system, not another seat license.

The installed system runs in your cloud account, under your keys. You pay model, cloud, API, and software providers directly. Foundry works through delegated access you can revoke, then operates the system or hands it over.

Talk through the architecture
  • 01
    Code and workflows

    The deployed system, operating skills, connectors, and release history.

  • 02
    Company context

    Data, memory, configuration, and the rules that govern access.

  • 03
    Credentials and accounts

    Your provider relationships and secrets remain under your control.

  • 04
    Work records and artifacts

    Decisions, logs, files, code changes, and outputs stay with the company.

  • 05
    The option to change operators

    Revoke Foundry's role or use the runbook to operate the system yourself.

How the engagement works

Target: one production workflow in 30 business days.

For a scoped implementation with access and integration readiness, each stage leaves a usable artifact in your environment. The 30-day window is a target, not a black-box promise.

Days 1–5

Specify

Map the problem, systems, permissions, owner, approval points, and definition of production-ready.

Days 6–20

Install

Stand up your environment, approved integrations, Company Brain context, and the first workflow.

Days 21–30

Govern

Evaluate, test approvals, release with monitoring and rollback, and leave an operating backlog.

After launch

Operate or hand over

Foundry monitors and improves the workflow, or your team takes the runbook and operates it.

Built from working software

We run the model ourselves.

Sky is Foundry 41's internal reference implementation. A messaging thread starts a job in a dedicated sandbox with controlled access to code and capabilities. It selects approved context, performs the work, and reports back with reviewable outputs.

Selected outputs—memory, code changes, files, todos, and routines—can survive a session. Sky and Grow, a second internal instance, are not customer deployments. Broader work identity and automatic resume remain architectural direction, not a production claim.

Ask to see it work

Bring one operational problem

Give important work an owned place to run.

Show us a workflow that crosses people and systems. We will map the context, approvals, integrations, and production path—and tell you whether it fits a 30-business-day implementation.