IThe economic shift
Syntax is cheap. Judgment is the bottleneck.
As AI lowers the cost of routine code synthesis, the constraint moves to problem discovery, handoff and translation latency, semantic verification, and decisions that last.
WATERFALL 1980sAGILE 2000s CONCERTED 2026+A NEW OPERATING MODEL / VOL. 01
THE THIRD ERA OF SOFTWARE ENGINEERING
Concerted
Engineering.
The AI-native software engineering
operating model.
Beyond shift left. Eliminate handoff latency through continuous, contract-mediated coordination.
Read the foundational paper COORDINATION ARCHITECTUREFIG. 01
Intent·Execution·Evidence·Durable ADRs
Shared context. Independent authority.02 / 02
THE SHIFT
From passing work downstream
to moving with a shared state.
+ THE CANONICAL DEFINITION
Concerted Engineering replaces repeated, document-mediated organizational handoffs with a continuously coordinated system of customer intent, bounded autonomous execution, independent verification, and operational feedback—while preserving explicit human accountability.
ORIGINATED BYDr. Joel He · OpenKedgeFOUNDATIONAL RESEARCH, 2026
01THE FOUNDATIONS
A different constraint.
A different way to engineer.
When implementation accelerates, coordination becomes the work. Three axioms define the operating model.
IIThe operating posture
Design the environment. Don’t push tasks.
Leadership moves from assigning sprint tickets and tracking story points to designing governed execution environments. Give people and agents clear priorities, boundaries, and room to act.
IIITopological safety
Independent verification is non-negotiable.
The entity that generates code—an AI agent or a human developer—must never approve its own production admission. Independence belongs in the architecture, not just the checklist.
THE JAZZ ENSEMBLE PRINCIPLEA shared lead sheet.
A shared lead sheet.
No central conductor.
A symphony orchestra follows a conductor’s baton. A jazz ensemble coordinates through a shared lead sheet: key, tempo, and harmonic invariants.
Here, the lead sheet is the shared engineering state. Intent, Execution, and Evidence Contracts let Product, Engineering with coding agents, and Verification specialists work asynchronously or in parallel, with strict separation of duties. No central conductor pushes every task; humans remain accountable for the boundaries.
02THE SHARED ENGINEERING STATE
Three contracts.
One version of intent.
Customer intent, execution boundaries, and independent proof stay synchronized as views of the same versioned engineering context.
01 / INTENT CONTRACT
What must be true?
Turn authentic customer evidence into explicit outcomes, behavioral requirements, and system invariants. A 90-day value hypothesis gives every change a reason to exist.
ACCOUNTABLE OWNERProduct & technical leads
- +Customer evidence
- +Behavioral requirements
- +System invariants
- +90-day value hypothesis
replica-rebalance@v1.3
Three synchronized views of one versioned context. A revision mismatch blocks admission.
intent.contract.yaml
# Illustrative schema · not an executable policykind: IntentContractcontext: replica-rebalance@v1.3owner: product-and-technical-leads initiative: read-replica-rebalancing-under-surgecustomer_evidence: source: support-and-runtime-telemetry trigger: p99_read_ms > 200 for 5moutcome: horizon_days: 90 hypothesis: reduce-surge-read-latencyinvariants: p99_read_ms_max: 200 acknowledged_write_loss: 0 residency: eu-west-1behavior: rebalance_reads: within-approved-replicasdecision_record: ADR-0042@v1Autonomy inside the boundary. Authority outside it. Code generators never approve their own production admission.
AXIOM III03THE SIX-PHASE PROCESS
A continuous loop.
Not another assembly line.
Six connected phases turn customer evidence into verified outcomes—and bring what you learn back to the beginning.
CONTINUOUS
COORDINATIONHuman accountability
at every phase.
COORDINATIONHuman accountability
at every phase.
Outcomes return to discovery. The loop stays open.
PHASE 01OF 06
Continuous discovery
Product + customer signalsAI continuously clusters user friction, support logs, and runtime telemetry. Human Product Managers validate the evidence and decide whether the problem is worth solving.
OUTPUT / SHARED STATEValidated customer problem
Next: Intent definition
FOLLOW AN INITIATIVE THROUGH THE LOOPDatabase read-replica rebalancing under surgeIllustrative scenario
An architectural illustration with synthetic values, not a measured deployment or an empirical result.
- 01
Continuous discovery
Support reports and query telemetry expose read-latency spikes during regional traffic surges. Product validates the impact with affected customers.
- 02
Intent definition
Intent v1.3 sets a p99 ceiling of 200 ms, zero acknowledged-write loss, and EU data residency. ADR-0042 records why read-weight rebalancing is the bounded intervention.
- 03
Bounded autonomous execution
An engineer initiates a MicroVM run. The agent can alter read-router logic and tests, but cannot drop schemas, access production credentials, or change primary-write routing.
- 04
Independent verification
A separately owned harness checks the 92% mutation threshold, zero-loss properties, interface compatibility, security findings, and residency constraints against the exact candidate revision.
- 05
Risk-aware delivery
The control plane checks signatures and contract versions before 1%, 5%, 25%, and 100% rollout rings. Excessive error-budget burn automatically stops and rolls back the canary.
- 06
Customer outcomes
In this illustrative scenario, p99 read latency falls 42%, from 310 ms to about 180 ms. The 90-day review checks sustained value, safety, and coordination cost before expanding the pilot.
42%illustrative p99 latency reduction
310 ms → ~180 ms, with the same zero-data-loss and residency constraints. Validate this hypothesis in your own environment.
04WHY NOW / THE PARADIGM SHIFT
The tools changed.
The operating model must follow.
Waterfall organized phases. Agile shortened iterations. Concerted Engineering proposes continuous coordination around shared state.
| OPERATING DIMENSION | ERA 01 / 1980s–1990sWaterfall | ERA 02 / 2000s–2010sAgile / Scrum | ERA 03 / 2026+Concerted Engineering |
|---|---|---|---|
| Execution & coordination | Sequential project phases | Iterative sprint batches | Continuous, contract-mediated loop |
| Role interaction | Siloed handoffs | Ceremonies & shift left | Continuous shared state |
| Feedback horizon | 12–18 months | 2 weeks | Hours to days |
| Routine implementation cost | High | High | Substantially reduced with AI |
| Primary bottleneck | Specification obsolescence | Local sprint optimization & drift | Handoff, translation, decision latency & semantic verification |
| Governance mechanics | Phase gates | Backlogs & standups | Governed autonomy through three contracts |
THE COST OF BOLTING AI ONTO OLD WORKFLOWS
01—06Six failure modes to recognize.
Review collapse
Generated code outpaces independent review. PR queues grow while verified delivery stalls.
Governance drag
Central review boards become chokepoints when every small change requires a meeting.
Internal sprawl
Decoupled teams duplicate shadow tools, policies, and infrastructure without shared boundaries.
Commitment volatility
Roadmaps reset under the assumption that AI can rewrite anything at any time.
Synthetic toil / tokenmaxxing
Token counts and PR volume become targets. Goodhart’s Law rewards activity over customer value.
Executive whiplash
Every model release triggers another reactive migration before previous decisions can produce evidence.
EMPIRICAL GROUNDINGEvidence for the question.
Evidence for the question.
Not a claim that it is settled.
These findings motivate a testable operating model. They do not constitute validation of Concerted Engineering.
The system determines the return.
DORA’s 2024 findings connected AI adoption with delivery tradeoffs. Its 2025 research frames AI as an amplifier of existing organizational strengths and weaknesses. These are contextual associations, not a causal test of Concerted Engineering.
Measure outcomes, not perceived speed.
The July 2025 randomized trial found experienced open-source developers took 19% longer with the tools studied. The February 2026 update reports selection effects and unreliable estimates of current uplift. Neither result establishes a universal AI productivity effect.
Code volume is an incomplete signal.
GitClear’s observational analysis of changed code raises concerns about churn and duplication. It motivates measuring durable code and maintenance cost; it does not isolate AI as the cause of every observed trend.
Standing on giants’ shoulders.
An extension of enduring ideas about flow, feedback, and independent proof.
Goldratt & Jevons
The Theory of Constraints directs attention to the system bottleneck. Jevons’ efficiency paradox suggests why cheaper implementation can increase demand—and move pressure downstream.
Theory of Constraints The Coal QuestionAgile & DevOps
The 2001 Agile Manifesto and DevOps established iterative adaptation and continuous delivery. This model extends that lineage by removing document-mediated handoff boundaries.
The Agile ManifestoMills’ Cleanroom
Harlan Mills’ Cleanroom work in the 1980s separated formal specification from independent statistical certification. AI-assisted execution invites us to revisit that separation at a faster cadence.
Cleanroom foundationsGlushkov’s OGAS, 1962
Benjamin Peters’ How Not to Network a Nation shows how institutional competition obstructed coordinated feedback. The implication here: shared state needs aligned authority and incentives.
Peters’ historical account05HOW YOUR ROLE EVOLVES
Less translation.
More engineering judgment.
Accountability gets clearer as execution becomes more autonomous. Select your role to see what changes.
FROM
TO / PRODUCTStatic 40-page PRDs and endless backlog grooming.
Author the intent. Own the outcome.
Validate authentic customer friction, express behavioral requirements in Intent Contracts, and stay accountable for 90-day outcome realization.
- Evidence-backed Intent Contracts
- Customer value hypotheses
- 90-day outcome reviews
Are we solving a problem customers actually have?
Human accountability is an invariant, across every role.
06THE 90-DAY ADOPTION EXPERIMENT
Start bounded.
Earn the right to scale.
Choose a real customer problem, one or two bounded services, and explicit stop-and-refine guardrails.
Download the adoption blueprintDAYS01—3001 / INTENT
Baseline & intent
Instrument handoff and translation latency. Retire token and PR-volume vanity metrics. Choose one or two bounded services and write customer-backed Intent Contracts.
THE DELIVERABLE
A measured baseline and an agreed value hypothesis.
Decision guardrail
Proceed when the problem, accountable owner, invariants, and outcome measures are explicit.
DAYS31—6002 / EXECUTION
Bounded execution
Provision isolated MicroVMs and policy-as-code rails. Encode permitted mutations, network scopes, and budgets. Let engineers pull stable priorities and orchestrate autonomous coding agents.
THE DELIVERABLE
A working sandbox with inspectable execution boundaries.
Decision guardrail
Pause if scope escapes, coordination cost grows, or ownership becomes ambiguous.
DAYS61—9003 / EVIDENCE
Evidence & decision durability
Enforce independent Evidence Contracts through an out-of-band control plane. Preserve versioned ADRs and compare realized customer outcomes with the original intent.
THE DELIVERABLE
An evidence-backed decision to expand, refine, or stop.
Decision guardrail
Expand only if customer value and flow improve without degrading safety or engineering health.
THE LEADERSHIP SCORECARDMeasure what
Measure what
survives the sprint.
Five dimensions keep local speed aligned with customer value and system health.
OUTCOMES OVER OUTPUT VOLUME01Customer value90-day outcome realization+
Compare realized customer outcomes with the original Intent Contract.
02Engineering flowIntent-to-production lead time+
Track total lead time and the share lost to handoff and translation.
03Organizational efficiencyVerified value / coordination overhead+
Measure outcome delivery relative to coordination time and cost.
04Engineering healthDurability, mutation score & stability+
Review 2-week code survival, mutation scores, and error-budget stability.
05Leadership efficacyPriority stability & ADR durability+
Track unplanned priority resets and the lifetime of accepted decisions.
07AN OPEN RESEARCH AGENDA
A TESTABLE PROPOSITIONThe next era is
The next era is
an open question.
“When product, engineering, and verification disciplines coordinate around a continuously updated state of customer intent, bounded execution, and independent evidence, how much handoff latency, translation overhead, and rework can be eliminated without compromising topological safety?”
We invite industry and academic collaborators to test the model, report boundary conditions, and share what fails as well as what works.
Request an executive briefingBUILD ON THE RESEARCHRead it. Test it.
Read it. Test it.
Cite it.
Dr. Joel He · 2026
OpenKedge Insights
@misc{he2026concerted,
author = {He, Joel},
title = {Beyond Agile: Concerted Engineering—The AI-Native Software Engineering Operating Model},
year = {2026},
howpublished = {OpenKedge Insights},
url = {https://www.openkedge.io/blog/beyond-agile-ai-native-software-engineering},
note = {Companion website: https://concertedengineering.com}
}