AI ENGINEERING CONTROL PLANE
Your AI developmentteam, managed.
Tell Foreman what needs to get done.
From intent to release, Foreman coordinates who builds, who independently verifies, what evidence is required, and when a human decision is needed.
AI can write code. Coordinating it is the hard part.
Teams now run a dozen coding agents across a dozen tabs. Every handoff is a human copying context between tools, re-checking output, and chasing what actually shipped. The work moves — but nobody can see the whole thing.
Today · manual relay
- 01Prompt
- 02Agent
- 03Result
- 04Copy
- 05Another agent
- 06QA
- 07Retry
- 08Slack
- 09Human
- 10Deploy
12+
tools in play
0
shared source of truth
∞
copy-paste handoffs
You shouldn't have to manage the AI team yourself.
Stop deciding which AI tool to prompt next. Foreman determines what capabilities the work requires, evaluates the resources available to perform it, and coordinates the evidence required before release.
- Without Foreman
- Which prompt do I write? Which agent goes first? Who reviews it? What still needs testing?
- With Foreman
- Tell Foreman what needs to get done.
Intent → requirements → eligible capability
Business intent
“I need a new customer onboarding flow.”
- Outcome
- Problem
- Ticket
- Requirement
- Brief
- Requirements shapingForeman work shaping, confirmed by a person
- UX / product designConnected design capability
- ImplementationEligible coding agent or developer
- Independent reviewA resource that did not author the work
- Test strategyEvidence requirements set per work item
- TestingIntegration, SIT, regression, and UAT as required
- ValidationDeterministic gates on the exact candidate
- Release authorityHuman authority
Define
Intent, requirements, evidence
Assign
Capability, eligibility, independence
Execute
AI or human capability does the work
Prove
Validation, review, testing, remediation
Release
Human authority
Bring the outcome. Bring the brief. Or bring the problem.
A detailed brief helps. It is not required. Foreman shapes intent into delivery-ready work: requirements, acceptance criteria, constraints, dependencies, architecture questions, risks, a definition of done, a test strategy, and the evidence the work must produce. People confirm the decisions that matter.
Foreman understands the capabilities available to delivery — AI agents, developers, reviewers, delivery leads, internal teams, and approved external or 3PL partners — and coordinates work around them. Human judgment and organizational authority stay explicit.
Scale software delivery before you scale headcount.
Before adding another developer, contractor, AI subscription, or delivery partner, understand how much capacity is already available — and how much is being lost to coordination.
Fragmented capacity
- Disconnected internal engineering
- AI tools operating independently
- Contractor / 3PL capacity
- Queues
- Rework
- Manual coordination
- Unused capability
FOREMAN
Define · Assign · Execute · Prove · Release
Governed capacity
- Structured demand
- Deliberate routing
- Visible workload
- Clearer dependencies
- Coordinated AI execution
- Targeted human judgment
- Evidence-driven delivery
Illustrative operating model — actual results depend on workflow, adoption and delivery environment.
One foreman. A crew of specialists.
Foreman sits above every agent you already use. Agents are interchangeable capabilities it assigns and governs — never the product itself, and never the judge of their own work.
Input
Business intent
A brief, an outcome, a ticket — in plain language.
FOREMAN
Orchestration control plane
Eligible agents · interchangeable capabilities
From assignment to release · builders never certify their own work
Implementation
01Candidate
The exact SHA produced by the assigned builder.
Required checks
02Deterministic validation
The candidate is tested — not the intent.
Separate reviewer
03Independent AI review
An agent that did not author the work, such as Codex, challenges the code and decisions, files findings, and confirms retests.
Your QA + security
04Independent verification
Your testing, security, and CI systems verify the same candidate and return findings.
Foreman
05Evidence
Results are collected, attached to the item, and checked against policy.
Human authority
06Human release
The decision waits for a person.
Every piece of work goes to an eligible execution path.
Foreman determines what capabilities the work requires, then evaluates the resources available to perform it. Intelligence informs the choice; it never overrides policy or authority.
Foreman works like a delivery lead for your AI crew — understanding what the work requires, which resources are available, who is independent, and what evidence is needed before the work can move forward.
Policy defines what is allowed. Intelligence helps choose among what is allowed.
- Capability
- Can this resource perform the work?
- Capacity
- How much work can it carry at once?
- Availability
- Is it free to take the work now?
- Eligibility
- Is it permitted for this work?
- Independence
- Is it independent from the author of the candidate?
- Performance history
- What does previous execution evidence show?
- Cost
- Is it consistent with the operating constraints?
- Policy
- Is the assignment allowed?
Work requirement
Independent review · candidate a52819de
WI-4821 · Enterprise SSO
| Resource | Capability | Eligibility | Independence | Performance history | Result |
|---|---|---|---|---|---|
| Claude Code | Pass: Code review | Pass: Cleared for repo | Fail: Authored candidate | Not evaluated: Not evaluated | Ineligible |
| OpenAI Codex | Pass: Code review | Pass: Cleared for repo | Pass: Did not author | Pass: Strong auth review history | Selected |
| Cursor | Pass: Code review | Pass: Cleared for repo | Pass: Did not author | Pass: Limited auth history | Eligible |
Claude Code
Ineligible- Capability
- Pass: Code review
- Eligibility
- Pass: Cleared for repo
- Independence
- Fail: Authored candidate
- Performance history
- Not evaluated: Not evaluated
OpenAI Codex
Selected- Capability
- Pass: Code review
- Eligibility
- Pass: Cleared for repo
- Independence
- Pass: Did not author
- Performance history
- Pass: Strong auth review history
Cursor
Eligible- Capability
- Pass: Code review
- Eligibility
- Pass: Cleared for repo
- Independence
- Pass: Did not author
- Performance history
- Pass: Limited auth history
Foreman can enforce separation between the capability that creates work and the capability responsible for independently reviewing it.
Illustrative Foreman workspace. Sample candidates and history shown for explanation, not as customer benchmarks.
Define. Assign. Execute. Prove. Release.
One operating model for every piece of work. Validation, review, and testing all live under Prove — and every stage leaves a record.
- D
Define
Business intent becomes structured, shaped engineering work.
Problems, requirements, tickets, and briefs are shaped into work items with constraints, acceptance criteria, a definition of done, and evidence requirements.
- A
Assign
Required capability is matched to an eligible resource.
Foreman weighs capability, eligibility, policy, availability, capacity, independence, performance history, and cost before assigning the work.
- E
Execute
AI agents or people carry out the approved work.
Implementation, design, research, remediation, and other approved execution work runs against the exact scope that was defined.
- P
Prove
Validation, independent review, and independent verification produce the evidence.
Deterministic validation, independent AI review, and your own testing, security, and CI systems — integration, regression, SIT, security, UAT, and smoke — inspect the candidate as the work requires. Foreman routes findings and tracks remediation.
- R
Release
A person makes the release decision with complete evidence.
Readiness, policy satisfaction, and unresolved findings are shown together. Nothing ships without an explicit human decision.
Not a dashboard. A system of record for AI-built software.
Foreman tracks the exact state of every work item — who built it, what was validated, what review found, and whether it is cleared for release.
WI-4821Enterprise customer portal
Awaiting releaseSelf-serve enterprise onboarding portal
Objective — Ship SSO-backed onboarding with role mapping and audit logging, validated against the enterprise test suite.
- Assigned provider
- Claude Code
- Work state
- Independent AI review
- Exact candidate
- a52819de
- Validation
- 143 / 143 passed
- Review findings
- 2 raised · 2 remediated
- Open findings
- 0
- Remediation
- Complete
- Evidence trail
- Complete
Human release decision required
Foreman never ships on its own.
Pipeline
- Define
- Assign
- Execute
- Prove
- Release
Other work items
Billing webhook retries
WI-4820 · Validating
Dashboard empty states
WI-4816 · Ready to release
SSO group mapping
WI-4809 · Assigned
Crew · live
- Claude CodeRemediating WI-4821
- CodexRunning 143 tests
- Independent AI reviewerConfirming retest
- GitHub CopilotIdle · ready
Illustrative product view.
Ask the questions a delivery lead asks all day.
Foreman understands the work, the available capabilities, execution state, evidence, and delivery constraints around it. Ask what is blocked, who should take the next task, what still needs testing, or what needs your decision.
Stop deciding which AI tool to prompt next. Ask Foreman what needs to happen next.
Bring the outcome. Bring the brief. Or bring the problem. Foreman helps shape intent into delivery-ready work — requirements, acceptance criteria, dependencies, constraints, test expectations, and a definition of done.
Delivery lead asks
Which AI capability should test this change?
Foreman Intelligence
Claude Code implemented the candidate, so it should not independently review its own work. Cursor is an eligible independent reviewer in this example. Before UAT, this change still needs SIT evidence and a smoke check.
CapabilityIndependent reviewTestingHuman releaseGrounded in WI-4821 · a52819deDelivery lead asks
What should the team work on next?
Foreman Intelligence
Clear the blocked authentication work first, then move the candidate awaiting independent AI review. One release is ready for a human decision.
BlockedAwaiting reviewRelease decisionGrounded in 3 work itemsDelivery lead asks
Why isn't this ready to release?
Foreman Intelligence
The candidate has passed implementation validation, but independent review and required test evidence are still incomplete.
Validation passedReview pendingEvidence missingGrounded in WI-4820 · evidence trail
What you can ask
- Resource decisions
- Who should do this work? Foreman weighs capability, availability, independence, policy, and delivery evidence.
- Delivery questions
- What is blocked? What is waiting? What should happen next?
- Testing strategy
- What test stage does this change need? Who can independently test it? What evidence is missing before UAT?
- Evidence-grounded explanation
- Why is this not ready? What evidence supports the recommendation?
- Current external intelligence
- When approved current public information is needed, Foreman can incorporate external research without handing the workflow to the research provider.
Evidence-grounded and workspace-scoped. Ask Foreman advises — it cannot approve, merge, deploy, or release on its own.
Illustrative product view. Providers shown are example capabilities, used when connected and eligible.
Testing strategy
Every change gets the evidence it actually needs.
Foreman determines what evidence and test stage the work requires and routes preparation or execution to eligible capabilities.
Define
- Intent
- Shaping
- Requirements
Assign
- Capability
- Eligibility
- Independence
Execute
- Build / implementation
- Remediation
Prove
- Deterministic validation
- Independent review
- Required evidence
- Unit
- Integration
- SIT
- Smoke
- UAT
- Regression
Release
- Human decision
Foreman coordinates the required testing stages across the capabilities available to the delivery workflow.
Builders build. Inspectors inspect. Foreman governs.
AI agents can implement the work. Independent systems can test, inspect, and challenge the result. Foreman coordinates both, collects the evidence, applies policy, and keeps release authority with your organization.
01 · Build
Builders
AI coding agents implement the work.
- Claude Code
- OpenAI Codex
- Muse
- GitHub Copilot
- Cursor
02 · Candidate
The exact work
The exact work being evaluated — pinned to its SHA.
Candidate SHA a52819de
03 · Review
Independent AI review
A separate agent that did not build the work reviews the implementation independently.
- OpenAI Codex
- Any agent that did not build it
04 · Verify
Independent verification
Your customer-controlled QA, security, and CI systems verify the candidate and return findings.
- BrowserStack
- Snyk
- mabl
- Tricentis
- QA Wolf
- Customer CI
05 · Govern
FOREMAN
Policy and decision gates across builders, reviewers, and inspectors.
- Collect evidence
- Apply policy
- Route findings
- Determine release readiness
06 · Decide
Human authority
A person with authority decides — approve, reject, or request remediation.
- Approve
- Reject
- Request remediation
Inspectors provide findings. Foreman applies policy. Humans retain release authority.
Builders never certify their own work, and no external inspector can approve a release. Foreman determines readiness under your policy — the release decision stays with a person.
Bring the systems you already trust
Keep the testing, security, and quality systems your organization already trusts. Foreman coordinates their results with AI-built work and applies your release policy.
For example BrowserStack, Snyk, mabl, Tricentis, QA Wolf, Customer CI. These remain your accounts, separately owned and billed by your organization.
Your code stays inside your walls.
Foreman's control plane orchestrates the work — but agents, repositories, and model keys live in your own environment. We coordinate the crew without ever holding your source.
With Private Runner, third-party provider credentials remain in your environment. Foreman receives verification results and evidence—not the provider credentials themselves.
Foreman control plane
Policy · routing · evidence
Your environment
Agent execution
Runs on your compute
Your repositories
Code never leaves
Your model keys
Providers you approve
Verification credentials
BrowserStack, Snyk, GitHub, and more
Independent verification · inside your boundary
- 01Foreman requests verification of the candidate
- 02Your Runner or CI uses your credentials
- 03Your provider verifies the candidate
- 04Results and evidence return to Foreman
Nothing ships without a paper trail.
Every work item carries a complete, immutable record: what was decided, which agent produced the candidate, exactly what was tested, and who signed off. When someone asks “how did this ship?”, the answer is one click away.
Brief created
WI-4821 · Enterprise Customer Portal
Assignment policy evaluated
3 eligible providers
Claude Code selected
capability + readiness match
Candidate produced
a52819de
Validation complete
143 / 143 tests passed
Independent AI review completed
2 findings raised
Remediation completed
candidate updated
Retest passed
143 / 143 tests passed
Release ready
evidence trail complete
Awaiting human approval
release authority required
Build performance history by capability.
Foreman records execution, validation, review and remediation outcomes so future assignments can be informed by actual delivery evidence.
Policy determines what is eligible. Performance history helps distinguish among the eligible choices.
Backend implementation
- Claude CodeStrongest history92
- Codex88
- Cursor81
UI implementation
- CursorStrongest history91
- GitHub Copilot84
- Claude Code79
The last decision is always yours.
Foreman coordinates the builders, the independent reviewers, and your verification systems — then it stops. Inspectors provide findings. Foreman applies policy. A release only happens when a person with authority approves it, with every piece of evidence in front of them.
- Agents can ask
- A builder may surface a question or a lean.
- Foreman can recommend
- An independent recommendation, with its reasoning and evidence.
- Your team decides
- The final call stays with a person who has authority.
- Candidate produced
- Passed
- Deterministic validation
- Passed
- Independent AI review
- Passed
- Functional verification
- Passed
- Security verification
- Passed
- Evidence complete
- Ready
- Human release approval
- Waiting
Inspectors report findings. They cannot approve a release.
One standard across every team you run.
Business units, workspaces, and projects inherit the same governance — isolated where they need to be, consistent everywhere.
Payments
Checkout
Ledger
Platform
Identity
Data
What could Foreman be worth in your organization?
Model your engineering capacity, external delivery spend, AI investment and coordination overhead — using your own numbers and assumptions.
Priced per team, scaled by active seats.
Every plan includes the full governed workflow. You only pay for seats that are actively building.
Team
Starting at
For a focused delivery team
Seats and terms confirmed during onboarding
- Structured briefs and controlled assignment
- Deterministic validation
- Independent QA with findings and retests
- Connect your existing AI crew on your own provider accounts
- Human release gates
Business
Starting at
For organizations coordinating multiple teams and AI resources
Seats and terms confirmed during onboarding
- Everything in Team
- Crew Intelligence and provider performance history
- Repository coverage tracking
- Private Runner
- Workspace isolation and audit trail
Enterprise
For regulated and large-scale delivery
Annual agreements with procurement support
- Everything in Business
- Private Runner and workspace isolation
- Role-based access and audit trail
- Custom onboarding
- Procurement and security review support
FORESIGHT · COORDINATION · HIGHER TOGETHER
A higher standard for building with AI.
Bring your own AI crew. Let Foreman plan, coordinate, validate, and get every release ready for your decision.