Function cockpit
Engineering
Agents build, test and document. Engineers own architecture and release.
Phase: Phase 3 — Functional orchestration
Run the intent-to-release graph as a governed system: decomposition, code generation, testing, documentation, vulnerability detection and deployment preparation are agent work; product intent, architecture and production release remain human.
Trust score
0
Touchless
0%
Human review
0%
Override rate
0%
Capacity released
25-40% engineering capacity in selected software-factory settings (case-specific)
Accountable human
Chief Technology Officer
CoS agent: Engineering Chief of Staff
Strategy, policy, observe, workflow, and AI GBS side by side
Five layers
What Engineering actually runs
A function is not one thing. It sets direction, writes the rules it is bound by, watches the world and itself, does the work on lines, and pushes repeatable transactions into a shared spine. Agents work in all five — not just the last two.
Strategy
16 agents
Agents that take positions, argue them, and get held to the outcome.
16 positions on the record
6 ratified, 4 in challenge
Policy
7 agents
The written rules every other agent is bound by, with version and owner.
34 binding rules
29 active, 5 superseded or retired
Observe
15 agents
Continuous sensing of the outside world and of the agents themselves.
15 watchers running
They watch the outside world and the agents themselves
Workflow
37 agents
Doing the work: lines, stations, and live units moving through them.
33 lines, 174 live units
23 held for a person right now
AI GBS
14 agents
Repeatable transaction processing run once for the whole enterprise.
4 shared towers
75 stations on this function run inside AI GBS
Strategy layer — positions on the record
Every position names its author, the agent paid to argue against it, and the conditions that would break it.
Investment Position
RatifiedHold the ceiling at A3 for portfolio & intake. Triage, estimation and dependency mapping are agentic. What the engineering organization spends its quarter on is a leadership commitment. Raise it only after two consecutive quarters where the override rate stays under five percent and every override has a written cause.
Portfolio & Intake Operating Position
DraftKeep the repeatable request volume in the spine and pull the exception tail back into the function. The spine is cheaper per unit; the tail is where the context that prevents rework actually sits.
Design Decision
RejectedHold the ceiling at A2 for architecture & design. Agents author the design options and the decision record. Architecture decisions constrain years of future work, so a review board ratifies them. Raise it only after two consecutive quarters where the override rate stays under five percent and every override has a written cause.
Architecture Strategy Position
In challengeKeep the repeatable design decision volume in the spine and pull the exception tail back into the function. The spine is cheaper per unit; the tail is where the context that prevents rework actually sits.
Architecture & Design Operating Position
DraftRedesign the lane around exception density rather than volume. Volume is already absorbed; the cost that remains is concentrated in a small share of design decision cases that no current agent resolves without a human.
Build Operating Position
DraftHold the ceiling at A4 for build. Code generation and dependency maintenance are near-touchless. A human still approves every merge into a protected branch — that gate is the whole control. Raise it only after two consecutive quarters where the override rate stays under five percent and every override has a written cause.
Policy layer — what binds the agents
Versioned rules with an owning agent, a human approver, and the agents they constrain.
Portfolio & Intake Decision Envelope
vv5.6Agents in this lane may act without a human when the request sits inside the stated value, risk and confidence envelope. Outside it, the unit holds at a gate with a named approver and a running clock.
Portfolio & Intake Evidence Standard
vv1.0Every autonomous decision writes inputs, the rule version applied, the model and prompt version, the output and a reversal path. An action with no evidence record is treated as a control failure, not a fast decision.
Portfolio & Intake Escalation Rule
vv3.5Escalation is mandatory when confidence falls below the floor, when two options sit inside the confidence band, or when a break condition on a ratified position fires. VP Engineering Operations owns the response clock.
Portfolio & Intake Autonomy Ceiling
vv3.8The ceiling for this lane is 74 on the platform scale. Triage, estimation and dependency mapping are agentic. What the engineering organization spends its quarter on is a leadership commitment. The ceiling is a governance decision, not a technical limit, and only VP Engineering Operations can propose moving it.
Architecture & Design Decision Envelope
vv5.8Agents in this lane may act without a human when the design decision sits inside the stated value, risk and confidence envelope. Outside it, the unit holds at a gate with a named approver and a running clock.
Architecture & Design Evidence Standard
vv1.6Every autonomous decision writes inputs, the rule version applied, the model and prompt version, the output and a reversal path. An action with no evidence record is treated as a control failure, not a fast decision.
Workflow layer — the lines doing the work
Each line is a workflow. Each station is a stage. Each unit is a live piece of work.
| Line | Sub-function | Kind | Units / wk | Touchless | Timetable | Actual | Stations |
|---|---|---|---|---|---|---|---|
| EN1A Intake-to-Backlog | Portfolio & Intake | transactional | 4,388 | 72% | 8.2h | 8.3h | 7 |
| EN1B Capacity Rebalance | Portfolio & Intake | analytical | 73 | 47% | 135.6h | 131.7h | 4 |
| EN1C Investment Position | Portfolio & Intake | judgment | 1 | 42% | 297.1h | 327.3h | 5 |
| EN2A Design Decision | Architecture & Design | judgment | 5 | 28% | 828.0h | 665.3h | 7 |
| EN2B Pattern Compliance Sweep | Architecture & Design | transactional | 6,939 | 67% | 6.6h | 8.3h | 5 |
| EN2C Architecture Strategy Position | Architecture & Design | judgment | 2 | 34% | 281.2h | 306.4h | 5 |
| EN3A Feature Build | Build | transactional | 1,610 | 81% | 18.7h | 24.5h | 6 |
| EN3B Dependency Update | Build | transactional | 6,826 | 82% | 22.8h | 24.6h | 6 |
| EN3C Documentation Sync | Build | transactional | 7,091 | 89% | 16.9h | 20.3h | 4 |
| EN4A Pull Request Review | Code Review & Quality | transactional | 6,838 | 89% | 11.9h | 14.8h | 6 |
| EN4B Quality Debt Sweep | Code Review & Quality | transactional | 304 | 86% | 19.1h | 22.1h | 4 |
| EN4C Convention Change | Code Review & Quality | transactional | 2,432 | 86% | 21.5h | 16.2h | 4 |
| EN5A Release Test Cycle | Test & Assurance | transactional | 5,597 | 89% | 1.7h | 2.2h | 6 |
| EN5B Flaky Test Remediation | Test & Assurance | transactional | 2,706 | 90% | 11.5h | 10.2h | 5 |
| EN5C Performance Baseline | Test & Assurance | transactional | 7,697 | 93% | 9.1h | 8.6h | 4 |
| EN6A Production Release | Release & Deployment | transactional | 2,598 | 79% | 23.0h | 18.7h | 7 |
| EN6B Emergency Revert | Release & Deployment | transactional | 6,809 | 88% | 10.9h | 9.5h | 5 |
| EN6C Release Readiness | Release & Deployment | transactional | 4,656 | 78% | 21.4h | 20.2h | 3 |
| EN7A Incident Response | Site Reliability | transactional | 7,793 | 85% | 17.7h | 17.8h | 8 |
| EN7B Error Budget Enforcement | Site Reliability | transactional | 3,053 | 85% | 24.4h | 31.5h | 5 |
| EN7C Capacity Response | Site Reliability | transactional | 5,263 | 85% | 23.6h | 20.6h | 4 |
| EN8A Debt Remediation | Technical Debt | transactional | 7,727 | 78% | 17.8h | 20.8h | 6 |
| EN8B Deprecation Campaign | Technical Debt | transactional | 4,772 | 82% | 6.1h | 7.3h | 5 |
| EN8C Debt Doctrine Position | Technical Debt | judgment | 4 | 49% | 268.4h | 345.6h | 5 |
| EN9A Environment Request | Developer Experience | transactional | 3,674 | 88% | 9.2h | 8.0h | 5 |
| EN9B Friction Remediation | Developer Experience | transactional | 5,890 | 94% | 22.4h | 25.3h | 6 |
| EN9C Developer Onboarding | Developer Experience | transactional | 4,391 | 90% | 9.5h | 9.2h | 5 |
| EN10A Vulnerability Remediation | Security Engineering | transactional | 3,326 | 81% | 1.0h | 0.9h | 7 |
| EN10B Secret Exposure Response | Security Engineering | transactional | 4,241 | 80% | 13.8h | 15.2h | 6 |
| EN10C Threat Model Review | Security Engineering | judgment | 3 | 52% | 655.1h | 708.3h | 6 |
| EN11A Monthly Engineering Read | Engineering Analytics | analytical | 83 | 75% | 168.1h | 208.6h | 6 |
| EN11B Cost per Service Read | Engineering Analytics | analytical | 52 | 62% | 171.7h | 142.8h | 4 |
| EN11C Measurement Doctrine Position | Engineering Analytics | judgment | 4 | 42% | 163.3h | 150.2h | 5 |
Observe layer — the watchers
Sensing agents and the agents that supervise other agents.
Engineering Intake Orchestrator
Owns the portfolio & intake lane end to end. Sequences the other agents, holds the timetable, and decides what surfaces to a human.
Portfolio & Intake Challenger Agent
Adversarial reviewer for portfolio & intake. Argues the opposite case on every position and flags where the evidence does not carry the claim.
Architecture Orchestrator
Owns the architecture & design lane end to end. Sequences the other agents, holds the timetable, and decides what surfaces to a human.
Architecture & Design Challenger Agent
Adversarial reviewer for architecture & design. Argues the opposite case on every position and flags where the evidence does not carry the claim.
Build Orchestrator
Owns the build lane end to end. Sequences the other agents, holds the timetable, and decides what surfaces to a human.
Build Challenger Agent
Adversarial reviewer for build. Argues the opposite case on every position and flags where the evidence does not carry the claim.
Code Quality Orchestrator
Owns the code review & quality lane end to end. Sequences the other agents, holds the timetable, and decides what surfaces to a human.
Complexity Monitor
Watches code review & quality continuously. Detects drift, quantifies it, and routes what matters without waiting for a reporting cycle.
AI GBS layer — the shared spine
Repeatable transaction processing does not belong to this function. It runs once for the enterprise.
Portfolio & Intake · Test & Assurance
Release & Deployment · Security Engineering
Site Reliability · Developer Experience
Engineering Analytics
Stations running inside AI GBS
75 of 176 stations(43% of the function's stations)
Read this page as the honest answer to “what would an 80% agentic Engineering look like?” The strategy and policy layers are where the argument happens. The workflow and AI GBS layers are where the volume goes. The observe layer is the only reason the other four can be trusted.
Sub-function breakdown