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.
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.
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.
DevCortex is a SaaS platform comprising:
dc_get_requirement, dc_get_backlog, dc_create_issue, dc_update_status and other tools to any MCP-compatible AI agentThe 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 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.
| 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 |
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.
Each sprint was implemented using Claude Code, connected to the DevCortex MCP server. The workflow followed a strict one-requirement-at-a-time protocol:
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 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.
Following each sprint's implementation, structured UAT test plans were generated directly from the DevCortex requirement database. Each test plan included:
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.
dc_create_issue autonomously when it identified a schema defect — without prompting. Issue was linked, resolved, and verified in the same session.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.
For investor enquiries and early access: devcortexai.com