Case Study

DevCortex as the Structured Intelligence Layer
for AI-Driven Development

How DevCortex powered the end-to-end development of For The Record across 4 production sprints — with zero platform blockers, full traceability, and autonomous agentic issue management.

0 Blocking Issues
100 Requirements
4 Production Sprints
~300 Tokens / Retrieval
100% Acceptance Criteria Met

Purpose-built context for agentic engineering

DevCortex is a purpose-built structured intelligence layer for agentic software engineering that sits between human specification and AI-agent execution. Its core innovation is a precision MCP (Model Context Protocol) server that delivers just-in-time, high-signal context to AI coding agents — delivering product specification and requirements on demand, then tracking verification results and maintaining full traceability of produced code back to requirements.

It also ensures more efficient token and context usage, eliminating the context rot that plagues vibe-coded, chat-session-driven development.

This case study documents the use of DevCortex to build For The Record, a full-stack career journal SaaS application, from initial specification through four production sprints. This beta test confirmed DevCortex's core thesis: that structured, on-demand context delivery produces deterministic, traceable, production-quality outcomes from agentic AI coding tools.

"The smallest possible set of high-signal tokens that maximize the likelihood of some desired outcome."

DevCortex operationalises this principle as a product. Every requirement is stored as a structured, queryable record. Claude Code fetches exactly what it needs — one requirement, its acceptance criteria, and any open issues — at the moment it needs it. Nothing more. It then updates DevCortex with build status and test result.


The Problem DevCortex Solves

Agentic AI coding tools like Claude Code, Cursor, Devin, and others are transforming software development. But they share a fundamental constraint: context windows are finite, and as sessions grow, AI performance degrades.

Anthropic's own engineering team named this phenomenon context rot: as the number of tokens in the context window increases, the model's ability to accurately recall information from that context decreases. The result, in practice, is the fix-it loop — the agent fixes one thing, breaks another, and the developer spends as much time re-explaining context as they save from automation.

Existing tools address parts of this problem. AWS Kiro adds spec-to-code generation but is locked to its own IDE. Traycer adds AI ticket creation but lacks structured requirements. None provide a complete, agent-agnostic, spec-driven control plane with automatic traceability.

What DevCortex Is

DevCortex is a SaaS platform comprising:

The platform applies an agentic V-methodology: requirements flow left-to-right through specification, implementation, and issue tracking, then fold back through UAT, traceability verification, and tagging — with DevCortex as the connective tissue throughout.


For The Record — Application Overview

For The Record is a career journal SaaS application built for professionals who want to capture their work narrative, track goals, and build an evidence base for performance reviews.

Sprint Scope

Category Detail
Stack Next.js 14 + Supabase + Vercel + Tailwind CSS
Requirements 100 requirements, each containing a User Story with 3–8 Acceptance Criteria
Sprints 4 full production sprints under DevCortex agentic control — base app foundation, then progressively adding functional capability with UAT at completion of each sprint
Blocking Issues 0 blocking issues

How DevCortex Was Used

01 Specification and Requirements Authoring

Before each sprint, a versioned specification document was authored in DevCortex. Requirements were structured with:

Requirements can be imported via CSV using the dcx CLI, or entered in the Web UI manually or with the built-in AI assistant. 100 requirements across four sprints were managed with this process.

02 Agentic Build Sessions via MCP

Each sprint was implemented using Claude Code, connected to the DevCortex MCP server. The workflow followed a strict one-requirement-at-a-time protocol:

dcx — MCP context session
# Prompt 0: establish session context
tool: dc_health()
✔ MCP connection verified
 
tool: dc_get_backlog(sprint="sprint-3")
✔ 25 requirements confirmed
 
# Agent fetches only what it needs
tool: dc_get_requirement("REQ-042")
→ title: "NextAuth.js HTTP-Only Cookies"
→ status: IN_PROGRESS
→ acceptance_criteria: [5 items]
→ open_issues: 0
 
# Precision context — not full chat history
tokens used: 300 / 200k
signal ratio: high
A typical requirement retrieval consumed ~300 tokens. Delivering the same information via repeated chat context would have consumed 10–50× more tokens and degraded agent accuracy with each passing message.

03 Autonomous Issue Management

One of the most significant observations from the beta occurred during Sprint 3 development. When Claude Code encountered a missing database column in the Supabase schema during implementation, it did not simply fail or ask for help.

Instead, Claude autonomously called the DevCortex MCP tool dc_create_issue, creating a structured issue record linked to the specific requirement where the schema gap was identified. The issue included:

This behaviour — autonomous issue creation linked to a requirement, without human prompting — confirmed a core DevCortex design hypothesis: when AI agents have access to structured issue management tools via MCP, they use them appropriately as part of the development workflow. The issue was resolved in the same session and the requirement was marked VERIFIED.
This is what agentic V-methodology looks like in practice. The agent identified a defect, logged it against the requirement, resolved it, and continued — all within the structured DevCortex context.

04 UAT Test Plans from Specifications

Following each sprint's implementation, structured UAT test plans were generated directly from the DevCortex requirement database. Each test plan included:

05 Traceability Matrix

The DevCortex traceability matrix provided a continuous, auto-generated view of requirement coverage. At any point in the sprint, the matrix confirmed which requirements had:

This replaced the manual effort of maintaining a traceability spreadsheet — a task that typically consumes significant QA time on conventional projects — with an automated, always-current view derived from the requirements database.


Key Findings

Context Efficiency
AI agents retrieved requirements on-demand at ~300 tokens per call. No context rot observed across 4 sprints. No requirement re-explanation needed between sessions.
Deterministic Outcomes
Every sprint delivered production-ready code with 100% acceptance criteria met on first pass. No architectural rework required.
🤖
Autonomous Issue Mgmt
Claude used dc_create_issue autonomously when it identified a schema defect — without prompting. Issue was linked, resolved, and verified in the same session.
🛡️
Zero Platform Blockers
DevCortex introduced no blocking issues across the entire For The Record beta. MCP connectivity was stable. Requirement retrieval was reliable throughout.
🔗
Traceability Without Effort
A complete REQ → AC → Test traceability chain was maintained across all 100 requirements. No manual matrix maintenance was required.
🧪
UAT Quality
Structured test plans generated from requirements produced higher-quality test cases than ad-hoc testing. Defect detection was systematic and requirement-linked.

The Missing Intelligence Layer

The For The Record beta test programme confirmed that DevCortex delivers on its core promise: structured, traceable, production-quality AI-driven development with no context rot.

In a market where every developer team is adopting AI coding assistants but struggling with reliability, consistency, and governance, DevCortex provides the missing layer. It is not an AI coding tool — it is the intelligence infrastructure that makes AI coding tools work predictably at scale.

There is no other product on the market today that provides a full agentic V-methodology MCP server platform with structured requirements, traceability matrix, and requirement-linked issue management — and works with any MCP-enabled coding agent. DevCortex occupies a unique position in the emerging agentic software engineering landscape.

"DevCortex is a structured intelligence layer for AI-driven development — proven in real end-to-end application development, not by benchmarks."

For investor enquiries and early access: devcortexai.com