# Concerted Engineering: a 90-day adoption experiment

Originated by Dr. Joel He · OpenKedge · 2026
Companion website: https://concertedengineering.com
Foundational paper: https://www.openkedge.io/blog/beyond-agile-ai-native-software-engineering

## Set up the experiment

Choose one or two bounded services and an authentic customer problem. Name accountable Product, Engineering, Verification, and Operations owners. Keep verification independent from code generation and production admission outside the agent execution path.

Before starting, record customer outcomes, intent-to-production lead time, time spent waiting at handoffs, translation and coordination time, two-week code survival, mutation score, and error-budget stability. Record the measurement window and service scope so later comparisons have a meaningful baseline.

## Days 1–30: baseline and Intent Contracts

- Instrument handoff, translation, and decision latency.
- Retire token volume, PR volume, and lines of code as productivity targets.
- Validate customer evidence and define a 90-day value hypothesis.
- Specify behavioral requirements, safety invariants, and ownership.
- Publish versioned Intent Contracts and immutable ADRs.

Deliverable: a measured baseline and an agreed value hypothesis.
Gate: proceed when the problem, accountable owner, invariants, and outcome measures are explicit.

## Days 31–60: bounded Execution Contracts

- Provision disposable MicroVMs or equivalently isolated execution environments.
- Define allowed repository paths, API scopes, and state mutations.
- Deny network access by default; allow only named dependencies.
- Set token, compute, and elapsed-time budgets.
- Keep production credentials and verification policy outside agent authority.
- Let engineers pull stable priorities and orchestrate coding agents.
- Preserve traces and exact contract and candidate revision references.

Deliverable: a working sandbox with inspectable execution boundaries.
Gate: pause if scope escapes, coordination cost grows, or ownership becomes ambiguous.

## Days 61–90: independent Evidence Contracts and decision durability

- Run separately owned conformance, mutation, property, security, and residency checks.
- Bind signed evidence to the exact Intent, Execution, and candidate versions.
- Let a separate control plane evaluate admission under human-accountable policy.
- Roll out progressively with explicit error-budget and rollback conditions.
- Preserve ADR versions; supersede accepted records explicitly rather than editing history.
- Compare realized customer outcomes and coordination cost against the baseline.

Deliverable: an evidence-backed decision to expand, refine, or stop.
Gate: expand only if customer value and flow improve without degrading safety or engineering health. Define numerical thresholds appropriate to the service before the experiment begins.

## Five-dimensional leadership scorecard

1. Customer value: 90-day outcome realization versus the Intent Contract.
2. Engineering flow: intent-to-production lead time and handoff latency reduction.
3. Organizational efficiency: verified outcome delivery relative to coordination time and cost.
4. Engineering health: two-week code survival, mutation score, and error-budget stability.
5. Leadership efficacy: priority stability and ADR durability.

## Record the outcome

Report scope, baseline, intervention, observation window, outcomes, safety failures, and limitations. Account for changing traffic, team composition, and model/tool versions. Separate measured results from illustrative architecture scenarios. Share negative findings and boundary conditions alongside successful outcomes.

This is an adoption experiment and a research proposition. Its thresholds must be validated in your environment; the companion site's example contracts describe an illustrative schema, not an executable policy engine.