Skip to content
In-Haus
Why In-Haus

Agile cannot manage machine-speed execution.

The prerequisite for autonomous execution is The Target Operating Model for Agentic Engineering.

Legacy Agile was built to manage the pace of human communication. Unregulated vibe coding drops management altogether. Under agentic execution, both leave the enterprise with runaway compute, architectural drift and code no accountable person has read.

The enemy

Two ways enterprises run AI-Built software today. Both fail at machine speed.

Legacy Agile and SAFe

Built for human communication latency.

Standups, tacit knowledge and working software over comprehensive documentation manage how fast people talk to each other. AI Build Agents cannot take part in those rituals. They act on what is written down, and Agile writes down too little.

Unregulated vibe coding

Built on conversational guesswork.

Developers prompt loosely and accept what comes back. Over continuous iterations, agents lose the architectural plot, act beyond their mandate and compound technical debt faster than engineers can review it.

Agentic governance

The 5 C's of Agentic Governance

Each C names a structural risk of unmanaged AI and the control In-Haus runs against it.

The CUnmanaged AIThe Target Operating Model for Agentic Engineering
CostMaterials BudgetAI spend without a ceiling is an unmanaged liability.Research: infinite execution loopsEvery build phase runs inside a fixed Materials Budget, with spend and forecast-to-complete on the record.
ControlSite LogisticsAutonomous agents without boundaries act beyond their mandate.Research: runaway autonomous executionSite Logistics holds every agent to its approved Build Spec and the Build Boundaries set for it.
CoherenceStructural IntegrityAgents that build from conversation build drift into the architecture.Research: context driftStructural Integrity: agents build against the approved Master Blueprint and nothing else, each Build Spec with its Inspection Checklist.
ComplianceCode ComplianceAI-built code must meet the same standard as any production system.Research: vulnerable AI outputCode Compliance: every Build Spec is held to the Building Code before it earns a Green Tag.
CommunityBuild ManagersAutomation without accountable people leaves the enterprise exposed.Research: automation biasNamed people govern every gate: the enterprise's Build Managers review, its testers approve, its leaders sign.
The case

Seven assumptions legacy methods rest on, and what agentic execution does to each.

Agile and SAFe were built for a world where writing code was the slow, expensive part. Under agentic execution, decisions, review and rework are the cost. The methods have not changed.

Legacy assumption
Under agentic execution
What breaks
The In-Haus control
Legacy assumptionBuilding is slow; plan it precisely.
Under agentic executionWriting code is cheap. Review and rework are most of the cost.Research: review loop overhead
What breaksEstimates miss the review loops, and work gets built twice.
The In-Haus controlCostEvery Build Phase is held to a fixed Materials Budget. Work is planned to be built once, and the spend is on the record.
Legacy assumptionOwnership is a handover at the end.
Under agentic executionAccounts, data and rules accumulate from day one.
What breaksThe Owner does not control the system it paid for.
The In-Haus controlControlThe enterprise owns its accounts, data and record from day one, and its code and database at Deliver.
Legacy assumptionTest at the end.
Under agentic executionCode arrives faster than people can review it.Research: cognitive overload in review
What breaksUntested features pile up.
The In-Haus controlCoherenceNothing ships untested. Work is complete when the enterprise's own people pass it, not when an agent says so.
Legacy assumptionDocuments are written about the work.
Under agentic executionDocuments can be generated from it.
What breaksStatus reports are stale on arrival.
The In-Haus controlCoherenceEvery report reads from the record. Status is current, auditable and never compiled by hand.
Legacy assumptionControl change to protect the schedule.
Under agentic executionChange is cheap; drift is the risk.Research: contextual ambiguity
What breaksScope moves silently, at machine speed.
The In-Haus controlCoherenceNo change moves the scope until its cost is known and a named person signs it. The history is kept.
Legacy assumptionOne team works one codebase.
Under agentic executionMany agents and people work it at once.Research: overwritten invariants
What breaksUntested combinations ship.
The In-Haus controlCoherenceWork is tested together before release, and released on a named person's signature. Failures surface; they are never carried forward.
Legacy assumptionTeam capacity is the constraint.
Under agentic executionThe Owner's decisions are the constraint.Research: the validation bottleneck
What breaksBuilds stall on open questions, not on code.
The In-Haus controlCommunityEvery open decision has a named, accountable owner, and its cost to the build is visible to the enterprise.

An enterprise should stand on its own systems without permanent outside help.

How each control runs is shared in the enterprise review.

Operating principles

Eight principles. Every In-Haus Agent works under them.

The 10 House Rules put these principles into force on every build. The complete gate map, House Rules and controls are shared in the enterprise review.

  1. A human hand on every gate.

    AI Build Agents build. AI Build Managers review. Named, accountable people decide and sign.

  2. The enterprise owns the outcome.

    Its data, decisions and Haus Rules from day one. Its code and database at Deliver.

  3. Decisions are the work.

    Under agentic execution, the open decision is the critical path. Every one is visible and owned.

  4. A gap is a question, never an assumption.

    Agents do not guess at intent. Missing intent goes to the person accountable for it.

  5. Evidence, not status.

    Every claim traces to the record. The record shows readiness; it does not certify.

  6. Feedback becomes rules.

    The enterprise's feedback becomes standing rules every agent on the build works under.

  7. Permission before action.

    Nothing runs without the Owner's permission.

  8. Nothing breaks quietly.

    Every failure reaches a person. No run quietly does less than it should.

In-Haus will never

  • Decide for the Owner, or act before permission is given.
  • Expose a credential.
  • Let an agent approve its own work, or ship work that was not tested.
  • Carry a failure forward silently.
  • Promise delivery speed, or claim more than the record shows.
  • Keep what belongs to the enterprise. Its data, decisions, Haus Rules, reports and transferred code leave with it.
The binary choice

Adopt The Target Operating Model, or accept unmanaged AI.

There is no third option. Agents already run at machine speed; the decision is whether governance runs with them.

Accept unmanaged AI

Risk is found after it ships.

Agents build from conversation, spend without a ceiling and ship code no accountable person has read. The enterprise learns the cost after the money is spent and the exposure after the release.

Adopt The Target Operating Model for Agentic Engineering

The building code for Agentic Engineering.

Eight phases, seven signed gates, 10 House Rules and the 21 controls of The Standard. Every requirement is written down before execution, every agent works inside its Build Boundaries, every Build Phase is held to its Materials Budget and the Building Code, and every gate is signed by a named, accountable person. The enterprise keeps the velocity of autonomous AI and gains structural control.

Before an enterprise scales its AI compute, it must scale its governance.

Stop cobbling together developer tools. Adopt the Target Operating Model for Agentic Engineering. The complete gate map, House Rules and controls are shared in the enterprise review.

Establish the Target Operating Model