Methodology & Process
Our Agent-Centric Development Cycle (ACDC)
Eastgate's flavor of harness engineering: spec-driven, test-driven, human-led
Eastgate Software - German Engineering Standards. Enterprise-Grade Results.
Our Agent-Centric Development Cycle (ACDC): Eastgate's flavor of harness engineering: spec-driven, test-driven, human-led
The industry is converging on a new way to ship software: harness engineering, spec-driven design, agents under human review. We did not invent any of these patterns. ACDC (Agent-Centric Development Cycle) is Eastgate Software's flavor of them - the specific way we package and apply these emerging practices on client engagements. Feedforward guides steer AI before it acts. Feedback sensors validate after. Human approval gates ensure nothing ships without sign-off. The result: 30-50% faster delivery with fewer defects.
Introduction
Why Do Most AI-Assisted Teams Still Struggle with Quality?
AI coding assistants changed how fast we write code. They did not change how hard it is to write the right code. The bottleneck was never typing speed - it was the gap between what the business needs and what the developer understands.
When requirements live in chat history, AI produces unpredictable results. No audit trail, no shared understanding, no way to verify intent. Teams iterate faster but rework more - because the input was never precise enough.
Our answer is a lightweight specification layer between intent and implementation. Requirements are encoded as structured, behavior-first specifications with concrete examples before any code is written. AI agents build from those specs, a code-reviewer agent enforces standards, and humans hold the approval gates.
What ACDC is, and isn't
ACDC is not a framework we invented. It is Eastgate's flavor of an emerging industry pattern. The underlying ideas come from the wider field: harness engineering as Martin Fowler describes it and OpenAI applies it on their internal Codex team, Specification by Example (Gojko Adzic, Manning 2011), the open-source OpenSpec project (Fission AI, 33K+ stars), and the test-driven and human-in-the-loop disciplines that predate AI. What we contribute is the packaging - the specific way we combine these practices, the standards we enforce, and the harness we have refined across years of delivery for Siemens Mobility, FinTech platforms, and enterprise SaaS.
The Evidence
What Happens to Code Quality Without a Harness?
Independent benchmarks across 470 real GitHub PRs and 211M lines of production code show what AI-assisted development looks like when it runs without structure. This is AI without a harness.
The pattern: code quality degrades, technical debt explodes, security vulnerabilities multiply. The harness is what makes AI-generated code production-ready.
Sources: CodeRabbit (470 PRs), GitClear (211M lines, 5-year study).
The Shift
Why Doesn't 10× Individual Productivity Compound to 10× Company Value?
AI made every individual roughly 10× more productive at certain tasks. No company became 10× more valuable. The gap is organizational, not technological. Bolting Copilot onto an existing process does not change the system. The industry is moving toward institutional AI - process, agents, and harness redesigned together. ACDC is how we work in that mode at Eastgate.
| What Most Companies Do | What We Do at Eastgate |
|---|---|
| Bolted Copilot onto existing process | Redesigned the process around AI |
| Every dev has their own AI habits | Coordinated agents with defined roles |
| AI generates everything, nobody filters | Harness filters signal from noise |
| AI agrees with everyone | Code-reviewer agent says no |
| +14 to 26% productivity gains | Multiplicative, different scale entirely |
| Same team, faster typing | New team shape: architects + operators + agents |
"Agent = Model + Harness. The harness is everything that supports the model's operation."
Harness Engineering
Why Is This Called Harness Engineering?
An AI agent is a model plus a harness. The model provides capability. The harness provides structure, constraints, and verification. Without the harness, AI produces unpredictable results. With it, AI builds from blueprints. When something fails, the fix is never "try harder." The fix is: what guide is missing, and how do we make it enforceable?
Industry signal
February 2026How OpenAI ships software with Codex agents
OpenAI publishes this exact discipline under the name "harness engineering" - the metaphor borrowed from horse tack: reins, saddle, and bit channeling a powerful animal. On their internal Codex team:
>90%
of the Codex app's code was generated by Codex itself
4 to 8
parallel agents per engineer at any given time
~1M
lines shipped in 5 months with no manually written source
"Humans steer. Agents execute."
Sources: OpenAI: Harness engineering, How OpenAI uses Codex (PDF), Martin Fowler: Harness engineering for coding agent users.
Talk to our engineers
Want to see what your current harness looks like?
Walk through your gaps and quick wins with our engineering team.
Book a consultThe two halves of a harness
Feedforward Guides
Steer AI before it acts
Structured GIVEN/WHEN/THEN specifications
Domain context files and terminology rules
Architecture constraints and coding standards
Templates with output format requirements
Feedback Sensors
Validate after AI acts
Computational: Linters, type checkers, test suites
Inferential: AI code review, quality gate agents
Human: Approval gates, architectural review
Continuous: Executable specs run on every commit
The steering loop: When issues recur, the harness strengthens. AI produces output. Sensors detect a problem. A human identifies the root cause. A new guide or sensor is added. The harness gets smarter over time. This is how our ACDC harness improves continuously - not through better prompts, but through better structure.
How ACDC Works
How Is ACDC Structured?
Five layers from raw AI to production-ready software. Buying an AI model is layer 1. The four layers above it are what we engineer at Eastgate.
AI Models & Agents
Execution engine
The raw capability layer. Buying an AI model lives here. The four layers above it are what we engineer at Eastgate.
Harness Engineering
Control system
Feedforward guides steer AI before it acts. Feedback sensors validate after. Hooks and rules enforce process at the boundaries.
Spec-Driven Design
Contract layer
Requirements encoded as structured, behavior-first specifications before code is written. AI builds from specs, not ad-hoc prompts.
Test-Driven Development
Verification layer
Every requirement is illustrated with concrete examples that define done before implementation. Tests are derived from specs and run continuously.
Human-in-the-Loop
Direction, judgment, decisions
Humans own decisions and approval gates. AI proposes, humans approve. Corrections feed back into the harness over time.
What the stack delivers
How Agents Operate
What Does Daily Work in the Harness Look Like?
The harness gives each agent a defined role and a contract to enforce. Coordination keeps parallel work from colliding. Bias control keeps AI from saying yes to everything.
Coordination: agents as stations on an assembly line
Each agent owns specific files. The spec is the blueprint. The harness is the standards manual that keeps stations from colliding, so parallel execution stays safe.
| Assembly Line | ACDC Agent / Artifact | What It Produces |
|---|---|---|
| Blueprint | Spec (OpenSpec scenarios, Given/When/Then) | Single source of truth for every role |
| Station 1: Engine | Backend-engineer agent (API, services, DB) | Working backend + unit tests |
| Station 2: Body | Frontend-engineer agent (components, pages) | UI + integration to backend |
| Station 3: QC | Tester agent (E2E + integration tests) | Spec coverage, regression safety |
| Final inspection | Code-reviewer agent (security, performance) | Gate pass/fail before merge |
| Standards manual | Harness (hooks, rules, context, quality gates) | The enforcement layer |
Bias control: a code-reviewer that says no
AI tools are sycophants by default. The most valuable agent is not a yes-man, it is a disciplined no-man - a code-reviewer that interrogates the code instead of flattering the developer. Specs become enforceable contracts, and the reviewer blocks merges that don't comply.
BA → Spec
Acceptance criteria become contracts.
Engineer
Unit tests assert these conditions.
Reviewer
Blocks if compliance fails.
Banned in specs
appropriate · reasonable · user-friendly · intuitive · seamless · efficient · fast
Vague words force ambiguity. Each must be replaced with a measurable condition.
Team Composition
What Does an Eastgate ACDC Team Look Like?
Humans shrink from 8 to 10 down to 4 to 7 per squad. Leverage multiplies because agents do the volume work under harness control. The team shape is architects, operators, and agents.
Architects
Direct the system. Provide the why.
BA / Domain Expert
Business understanding, acceptance criteria, QA validation, ~80% of UI/UX wireframes. Client side or EGS side.
Solution Architect / Senior Engineer
Architecture, tech stack, security, DevOps, harness rules, OpenSpec scenarios. Best-practices enforcement.
Operators
Run the system. Validate output. Handle edge cases.
Mid-level engineers operating the harness with high leverage. They validate AI-generated code against specs, execute TDD refactoring, and triage bugs. The harness enforces the standards, not the person.
AI Agents
Execute within the harness. Unlimited parallelism.
Backend · Frontend · Tester · Code-reviewer (the no-man) · Docs · Orchestrator.
The labor signal: junior developer employment is down roughly 20% since 2024 (Stanford HAI 2026). The industry needs fewer code-writers and more code-evaluators. We built our ACDC team shape around that shift.
Impact
What Does Adopting ACDC Change?
The shift is felt on two sides: the business side that funds the program and the engineering team that runs the harness day to day.
For the business
Up to 55% faster task completion
AI-assisted coding cuts task time from 2h41m to 1h11m on benchmark tasks. On greenfield engagements we see spec to production in days, not weeks.
Bounded technical debt
AI without a harness produced 8× more 5+ line code clones across 2024. The harness's pre-merge sensors block that pattern before it lands.
56% of firms see engineering cost reductions
Software engineering shows the most consistent cost benefit from AI adoption. The harness is what makes those gains durable across releases.
Predictable delivery, not surprise bugs
AI without process discipline initially raises change failure rate. Specs as enforceable contracts plus a reviewer agent turn AI speed into measurable delivery.
Onboarding cut roughly in half
Engineers using AI daily hit their 10th PR in ~49 days vs ~91 days without. With the kit's standards pre-installed, the curve stays consistent across teams.
For the engineering team
Code-evaluators, not code-writers
Junior developer employment is down ~20% since 2024 as AI absorbs routine coding. Engineers shift to architectural judgment, validation, and edge-case handling.
4 to 8 parallel agents per engineer
Mirroring OpenAI's internal pattern. One person now ships what a small team used to, with the harness keeping outputs consistent and merges safe.
Code documentation in half the time
McKinsey measures ~50% time reduction on documentation, ~50% on writing new code, ~33% on refactoring. The harness lets seniors design and review, not police PRs.
35 to 39% speed-up for newer developers
Less experienced engineers see the largest gains from AI assistance. The harness adds the standards layer so juniors ship correct code, not just code.
Fewer production incidents
Generic AI review reduces code review time by ~3% but can raise change failure rate without process discipline. Sensors and the no-man reviewer block risky merges pre-ship.
Talk to our engineers
Want to see what these outcomes could look like for your roadmap?
30-minute scoping call. No commitment, no slide deck.
Book a consultOur Engineering Kit
What Does Our Harness Look Like in Practice?
The harness ships as an internal package, the EGS Engineering Kit, installed into every project as a development dependency. It bundles agents, hooks, rules, role pipelines, and knowledge packs into a single drop-in. New engineers inherit the standards on day one rather than learning them by osmosis.
AI Agents
Specialized agents for planning, building, testing, reviewing, debugging, documentation, and quality checks.
Lifecycle Hooks
Event triggers that enforce process at session, prompt, tool-use, and commit boundaries.
Coding Rules
Workflow rules and stack-aware coding standards that load automatically by file type.
Role Skills
Pipelines for pre-sales, BA, dev, QA, DevSecOps, and PM, each from input to deliverable.
Knowledge Packs
Curated guides our agents consult mid-task: engineering standards, security, NFRs, and stack-specific patterns.
Foundations: the kit is built on Claude Code and OpenSpec for spec-driven development. Stack rules auto-load by file type, role pipelines suggest the next step, and lifecycle hooks enforce process at the boundaries. The harness gets smarter every release because every correction lands in the package, not in a single team's head.
FAQ
Common Questions
How is ACDC different from AWS Kiro, Cursor, Copilot, or other AI-coding tools? +
ACDC is a configuration layer that sits on top of Claude Code - agents, hooks, rules, and role pipelines packaged for the full engineering lifecycle. Tools like AWS Kiro and Cursor are standalone IDEs you install and license. The choice is not "Claude vs Claude" - both can use the same models. The choice is "buy a turnkey IDE optimized for an individual developer" versus "shape an open methodology around how your team actually ships." For organizations with BAs, QAs, DevSecOps, and PMs in the loop, the second answer fits better.
Does ACDC cover roles beyond developers? +
Yes. ACDC ships role pipelines for the full SDLC: BA (brief, story, review, testplan, handoff), QA (testplan, bug-report, regression, perf-test), DevSecOps (threat-model, pipeline, deploy, rollback, incident), PM (backlog-groom, sprint-plan, standup, retro), Pre-Sales (discovery, mockup-brief, client-qa, estimate), and Solution Architects. Most AI-coding tools are developer-only. ACDC was designed around the assumption that the engineering team is more than its developers.
What about vendor lock-in? +
ACDC is portable by design. The harness ships as markdown rules, agent definitions, hook scripts, and OpenSpec specs - plain files in a git repository. There is no proprietary IDE, no required cloud account, and no licensed runtime. Specs belong to the client in version control. The ACDC layer adds value on top of Claude Code; if you ever need to fork, modify, or migrate to a different model, the specs and standards remain yours.
How is this different from traditional Agile? +
Agile ceremonies still apply. The difference is that AI has a precise, auditable input (structured specs with GIVEN/WHEN/THEN criteria) instead of ambiguous user stories. Teams collaborate on specifications before implementation, reducing rework caused by misunderstanding.
Does this slow down development? +
Upfront, writing specs takes more time than jumping into code. But the net effect is faster delivery because AI-generated code matches intent on the first pass. For small bug fixes, we skip the full spec process - the overhead only pays off for features and greenfield projects.
What is OpenSpec? +
OpenSpec is an open-source SDD framework (MIT license, 33,000+ GitHub stars) for AI coding assistants. It is tool-agnostic (works with Claude Code, Cursor, Copilot, and 20+ others), brownfield-first, and iterative. Its propose-apply-archive workflow aligns with how we structure client engagements.
How does human-in-the-loop work in practice? +
Every critical transition has an approval gate. A developer and stakeholder collaborate on a proposal. AI generates the spec draft. The team reviews. AI implements from the approved spec. A human verifies. Nothing ships without sign-off. Human corrections are captured in spec history, so AI improves over time.
Can this work with our existing workflow? +
Yes. It layers on top of your existing Git workflow, CI/CD, and project management. Specs are stored alongside code. Most teams adopt incrementally: start with one feature, measure results, then expand.
What is harness engineering and how does ACDC relate to it? +
Harness engineering is the emerging industry standard where the engineering team's primary job is building the system that enables AI agents to do reliable work - not writing code directly. The harness consists of feedforward guides (specifications that steer AI before it acts) and feedback sensors (tests and reviews that validate after AI acts). ACDC is our implementation of harness engineering: specs are the guides, tests are the sensors, and human-in-the-loop provides the approval gates.
What is the difference between AI-assisted and AI-first? +
AI-assisted means bolting AI onto your existing process - an engineer opens Copilot, a PM drafts specs with ChatGPT, QA experiments with AI test generation. The workflow stays the same. Efficiency goes up 10-20%. AI-first means redesigning your process, architecture, and quality gates around the assumption that AI is the primary builder. ACDC is AI-first: structured specs become executable artifacts that AI agents build from, with automated quality harnesses at every transition.
Read the Full White Paper
Detailed framework, implementation methodology, and actionable insights - available instantly with your business email.
About Eastgate Software
Eastgate Software is a strategic engineering partner headquartered in Hanoi, Vietnam, with offices in Aachen, Germany and Tokyo, Japan. With 200+ engineers, 93% team retention, and 12+ years of delivery excellence, we build mission-critical systems for clients including Siemens Mobility and Yunex Traffic.
Our ACDC (Agent-Centric Development Cycle) methodology combines German engineering discipline with Vietnamese engineering talent to deliver enterprise-grade results across Intelligent Transportation, FinTech, Retail, and Manufacturing.
Contact: [email protected] | (+84) 246.276.3566 | eastgate-software.com
Ready to Transform Your Engineering Process?
See how our ACDC harness engineering methodology can accelerate your delivery. Start with a 2-week pilot on a real project.
Engineers
ACDC (Agent-Centric Development Cycle)
Retention
Partners, not vendors
Years
Enterprise delivery