The Work

Define the Outcome. Staff It. Reach Every System It Touches.

What the business signs, who works it, where the work goes. One signed definition of done. One roster of people and agents. Reach into every system the work lives in, including the ones with no API.

Part one of three · The work · The control · The proof
DEFINE · WHAT DONE MEANSSTAFF · WHO WORKS ITREACH · EVERY SYSTEM IT TOUCHESDefinition of doneacceptance criteria · not a promptBacklog cleared to the agreed agemeasure · source · risksEvery case has a disposition on recordmeasure · source · risksExceptions routed to a named ownermeasure · source · risksEvidence pack accepted by auditmeasure · source · risksowner · risk · auditbefore anything runsSIGNEDBY THE BUSINESSSUPERVISORHOLDS THE RESULTresearch-02own identityunder contractclassify-01own identityunder contractJ. Alvarezown loginunder contractdraft-03own identityunder contractR. Chenown loginunder contractone shared working statehand-offs verified, never trustedDOOR 1An APIcontract allow list,enforced before the callsubmitstatusreadcancel · modify ✕ deniedDOOR 2Documentation onlyAriad reads the docs andbuilds the binding{ }scoped to this outcomeDOOR 3A console, no APIGoverned Workspace inyour own cluster> submit JOB0142ok · recordedevery keystroke recordedsigned before anything runsa definition of done, not a prompt or a decka roster, not a botagents and people, one supervisor, one recordthree doors to every systeman API, documentation only, or a console
What this part answers

What Are We Trying to Prove Done, and Who Does It?

Three answers, in the order the page tells them. Select one to see it up close.

Signed before anything moves

A definition of done the business signs. Not a prompt, a deck or an intention. The objective every agent is bound to.

Section 01 · Outcomes ↓
The first answer, up close

Nobody argues about done after the fact.

Before a single agent runs, the business signs what done means. Everything that follows is measured against that signature, and the proof is built as the work happens.

Section 01 · Outcomes

Start from the outcome. Not the ticket.

One portfolio of every outcome in flight, each criterion tied to the evidence behind it.

Go to Outcomes, section 01 →
Section 01 · Maps and deliverables

Set the standard once. Apply it a thousand times.

A map sets the shape; every deliverable inherits it. Progress is counted, never typed.

See maps and deliverables, section 01 →
The second answer, up close

Agents and people on one roster, under one supervisor.

Every agent has a name, a contract and a job. So does every person. And the one that checks the work is never the one that did it.

Section 02 · Workforce

No black boxes. Every agent accounted for.

What it did, why, under which rules and at what cost, down to each tool call.

Go to Workforce, section 02 →
Section 02 · Four roles

The validator is never the executor.

Planning, execution, validation, oversight: enforced by the platform, not a line in a prompt.

See the four roles, section 02 →
The third answer, up close

Door one is an API. Every vendor has that. Doors two and three are where the demos stop.

Your highest-consequence systems were never built for agents. Two doors nobody else opens get Ariad to them anyway, under the same contract, onto the same record.

Door two · documentation only

No connector? Not a blocker.

Hand Ariad the vendor's documentation. It builds the binding this outcome needs, and nothing more.

GET claims allowPOST note allowPUT status gateDELETE claim deny
See how the binding is built, 03 · A →
Door three · a console, no API

No API at all? A console will do.

A governed Workspace in your own cluster, where the contract decides what the agent may type.

$ logon TSO batchop
READY
$ submit 'BATCH.RECON.NIGHTLY'
JOB RECON01 SUBMITTED
contract submit, status, read spool · denied cancel, modify
$
See a Workspace session, 03 · B →

The page tells it in that order: 01 Outcomes · 02 Workforce · 03 Reach · 04 Integration control

The page tells it in that order: 01 Outcomes · 02 Workforce · 03 Reach · 04 Integration control

01Outcomes · what the business signs
Outcomes

Start from the outcomes you are accountable for. Not from tickets.

Every outcome in flight in one place: criteria met, approvals pending, the whole workforce. Each criterion links to its evidence, and anything waiting on a person arrives with a recommendation.

  • Define an outcome, or describe a problem and let Ariad propose candidates
  • Progress is n of N verified, never a percent typed by hand
In the product: Outcome Portfolio · Approval packets
Outcome Portfolio
Outcomes in flight7
Criteria met41 of 52
Pending approvals3
Workforce19 agents · 11 people
Client issue resolution, verified5 of 5 criteriaon track5 agents · 2 people
Branch servicing migration9 of 14 criteriagate pending9 agents · 4 people
Regulatory data remediation7 of 9 criteriaon track4 agents · 3 people
Define a known outcomeDescribe a problem; Ariad proposes candidate outcomes
start from the outcomes you are accountable for, not from tickets
Maps and Deliverables

The map sets the standard. Deliverables apply it at scale.

A map sets the phases, owners and expected evidence once. Each deliverable applies it to one real unit, a branch, a region, an application, with its own checklist.

  • Change the map once; every deliverable inherits the standard
  • The second outcome reuses the shape of the first
In the product: Outcome Maps · Deliverables
Maps and Deliverables
Outcome map: branch servicing migration
Phase 1 · inventoryPhase 2 · migratePhase 3 · cutover (gate)Phase 4 · verify
assignees · expected evidence · status
Dallas branch
24 of 24 verified
Phoenix branch
24 of 24 verified
Austin branch
17 of 24 verified
Denver branch
11 of 24 verified
Chicago branch
5 of 24 verified
Seattle branch
0 of 24 · not started
Outcome Maps · Deliverablesprogress is n of N verified, never a percent typed by hand
02Workforce · who does the work
Workforce

No black boxes. Every agent and person on the outcome, with what it did, why, under which rules, and at what cost.

A supervisor holds every agent to the result over one shared working state, with runs, errors, cost and health visible per agent down to each tool call. People sit on the same roster, in the same audit trail.

  • Four roles: planning, execution, validation, oversight. The validator is never the executor
  • Each agent has its own identity and a contract; hand-offs verified, never trusted
In the product: Workforce · Agent observability · Activity log
APLANNINGplans the workAEXECUTIONdoes the workAVALIDATIONchecks the workAOVERSIGHTholds the outcomeVERIFIED, NEVER TRUSTEDRULE: THE VALIDATOR IS NEVER THE EXECUTORMRJTPEOPLE ON THE SAME ROSTER · THEY HOLD THE ACCOUNTABILITY GATES · SAME AUDIT TRAIL
03Reach · the three doors, and the two nobody else opens
Reach

Your most important systems have no API.
Ours doesn't need one.

Mainframes, green screens, appliances, vendor products with nothing but documentation: the highest-consequence systems in the building, and where every other demo stops. Ariad reaches them under the same contract, onto the same record.

03 · ADoor two, up close · a binding built from documentation

The vendor has an API and no connector. Ariad reads the documentation and builds one.

Ariad reads the vendor's documentation and binds only the operations this outcome needs, each one allowed, gated or denied by the contract. Nothing else is bound.

Vendor documentation · claims productno connector exists
read · bind · scope
Binding · scoped to the outcome6 operations found · 3 allowed
GET/v2/claims/{id}allow
GET/v2/claims/{id}/historyallow
POST/v2/claims/{id}/notesallow
PUT/v2/claims/{id}/statusgate
DELETE/v2/claims/{id}deny
POST/v2/paymentsdeny · out of scope
contract: claims-resolution · rsh-04PII fields masked · every call signed

The binding runs inside your environment under the agent's own identity; every call is signed into the pack.

03 · BDoor three, up close · a governed Workspace

A console is enough. The contract is the allow list.

A Workspace is a governed sandbox in your own cluster where an agent reaches a system with no API, through a terminal, for instance. Not screen automation, which replays a recorded path and breaks when the screen moves: the agent gets an environment and a contract, and the contract bounds what it may do.

Workspace core-batch-01 · your clusteragent batch-operator · bat-04
$ logon TSO batchop
READY · workspace core-batch-01 · sandbox in your cluster1
$ submit 'BATCH.RECON.NIGHTLY'
JOB RECON01 SUBMITTED2
$ status RECON01
RECON01 ENDED RC=0000 · 00:04:12
$ cancel RECON02
BLOCKED · contract bat-04 · cancel not allowed · recorded3
$ release DOWNSTREAM.CHAIN
GATE · production release · waiting on J. Alvarez4
SESSION SIGNED · 14 commands · in the evidence pack5
$
What you are watching
1
An environment, not a recordingA terminal in a governed sandbox in your own cluster. Ariad provisions it for the outcome; the agent works inside it. No API needed.
2
Allowed commands runThe contract is the allow list: submit, status, read spool. Allowed commands reach the system and are recorded.
3
Denied commands never arrivecancel and modify are outside the contract. The command never reaches the mainframe, and the attempt is on the record.
4
A gate where policy says soReleasing downstream jobs waits on a named owner. Where policy requires, your people run the production steps.
5
The session is evidenceEvery keystroke captured and signed, in the pack with the rest of the outcome.
Screen automationReplays a recorded path. Breaks when the screen moves. No contract, no record.
A WorkspaceGives the agent an environment and a contract. The contract bounds what it may do, whatever path it takes.

Workspace images are certified with your platform team; where policy requires, your people execute the production steps at a gate.

04Door one · integration control
Integration Control

Each connected service, and exactly which of its tools each agent may call.

Allow and deny are decided at plan review and enforced for the life of the outcome. Connections are monitored, versioned and owned.

  • Least privilege per agent, per integration, per outcome
In the product: Integration Control · Connection lifecycle
Integration Control · ticketing service
ticketing · ticket.read · research-agent-02
ticketing · ticket.update · update-agent-04
ticketing · ticket.delete · all agents
cloud · console.admin · all agents
source control · pull_request.open · build-agent-01
monitoring · alert.ack · ops-agent-03
decided at plan review · enforced for the life of the outcome
Reaches what you already run

Every category, three doors, one rule: nothing is connected until a plan review grants it, tool by tool. The catalog lists each system by how it is reached.

Open the integrations catalog

Outcome delivered · attributed

Application onboarding at an F500 payment processor, run as one outcome under governed automation.

The process mapped, the systems named up front, the approvals built in as gates. The time moved because the work and the controls moved together.

120 → 10days to onboard an application. Use case: M&A Cloud Onboarding Acceleration, F500 Payment Processor.
The Scaffolding Map, the field guide behind these pages, is available to teams evaluating Ariad.Ask for the field guide
Questions operators ask

In Your Words.

Can we start with a single outcome?

Yes. Runtime Edition is a fixed set of outcomes built with you and handed over. Most clients start with the one they are measured on.

What if the system has no API?

A governed Workspace drives it through its console, every keystroke signed. The contract is the allow list.

Who signs the definition of done?

The business owner of the outcome. Risk and audit see it first. Our Outcome Desk drafts it with you; it is yours once signed.