From Ad-Hoc AI to an Enterprise Operating Model
What began as a GenDD training program for 180 people evolved into a full SDLC scaffolding engagement that produced 41 production-ready deliverables: an enterprise operating model for AI-augmented software development.
Two phases, one operating model for AI-augmented delivery
Vanco Payment Solutions engaged HatchWorks AI across two phases to transform how their engineering organization builds and ships software.
Phase one was a GenDD training program for 180 people. Phase two turned that shared methodology into a complete SDLC scaffolding system: the playbooks, templates, guardrails, and workflows that make Generative-Driven Development practical at enterprise scale.
The scaffolding engagement didn't produce a strategy deck. It produced a working operating system for AI-augmented software development: the VancoSDLC Framework.
AI was already accelerating tasks. Scaling it needed governance.
Vanco had already proven that generative AI could accelerate individual tasks. Engineers were using GitHub Copilot, and isolated experiments were showing promise.
But scaling AI across a complex engineering organization with multiple product lines, legacy systems, and distributed teams required something fundamentally different: consistency, governance, and repeatable workflows, not one-off prompt hacks.
What Vanco was up against
-
01
A fragmented technology landscape
The codebase was roughly 70% legacy and 30% modern, spanning .NET Framework, Web Forms, Python 2.7, GoLang, and significant business logic embedded in SQL stored procedures. Recent M&A activity (the ACS Technologies acquisition) added further fragmentation.
-
02
Tribal knowledge as the bottleneck
Architecture and system behavior lived in the heads of senior engineers, not in documentation. New developers needed months to become productive. Legacy codebases had little to no navigable documentation.
-
03
Upstream quality gaps cascading downstream
Fewer than 15% of user stories used the acceptance criteria field effectively. Requirements entered development as ambiguous bullet points, and QA became a downstream safety net, discovering requirements rather than validating them. An estimated 10 to 15% of active work was rework or reversions.
-
04
Ungoverned AI usage creating new risks
GitHub Copilot was accelerating code generation, but without governance it shifted bottlenecks downstream: inconsistent pull requests, increased review burden, and no way to ensure generated code matched Vanco's architectural patterns or security requirements.
-
05
Testing as a safety net, not a quality system
Test automation coverage was approximately 20%. QA teams were catching basic functional issues that should have been prevented upstream. Automation deferral was common with no tracking of reasons.
Jira data analysis confirmed the primary constraint was input quality, not engineering speed.
Train the organization first, then scaffold the system
Phase 1: GenDD training and workshop
Building fluency and demand across the organization
Before touching a single workflow or template, HatchWorks AI delivered a comprehensive GenDD training program to Vanco's engineering organization. The rationale was straightforward: scaffolding only works if teams understand the methodology it encodes.
attendees trained
sessions, 16 hours of instruction
role-specific tracks: developers, architects, product owners, QA and more
tools covered hands-on, including Cursor and N8n, across 4 practical projects
Training didn't just teach skills, it created demand for the scaffolding engagement. The Vanco team wasn't waiting to be told to adopt AI. They were asking for the tools and frameworks to do it properly. Training builds fluency, fluency creates pull, and scaffolding operationalizes the pull into a repeatable system.
Phase 2: SDLC scaffolding
Turning methodology into an enterprise operating model
With 180 trained practitioners ready to go, HatchWorks AI engaged with Vanco to design and implement a complete SDLC scaffolding system. The engagement followed a structured four-sprint methodology. Select a sprint to see what it produced.
Sprint 1: Diagnose
HatchWorks conducted targeted interviews with product stakeholders, engineers, architects, and QA representatives. The team performed deep repository reviews of representative codebases, mapped workflows from intake through deployment, and established maturity baselines across process, architecture, testing, and AI readiness dimensions.
Vanco's SDLC operated as a handoff-driven, governance-constrained system. While Scrum ceremonies and two-week sprints were in place, the system behaved closer to a stage-gated, approval-driven model than an adaptive delivery system.
Sprint 2: Prescribe
Using discovery findings, HatchWorks defined the target SDLC blueprint: a stage-by-stage transformation plan covering ideation, refinement, development, testing, release governance, and post-release improvement. Every stage received an AS-IS assessment, a TO-BE target state, explicit Human/AI boundary definitions, and mapped playbooks to operationalize the transition.
This phase also produced the story and requirements model, architecture standards, testing strategy, PR review guardrails, and the AI augmentation map.
Sprint 3: Implement
The team built the production-ready assets: the cursor-rules repository, code and test templates, architecture scaffolds, the Brownfield Analysis engine, Context Packs, and a reference implementation repository demonstrating the complete story-to-code-to-test golden path.
Sprint 4: Socialize
The final sprint focused on embedding the new system into the organization through hands-on enablement, clear ownership, and measurable adoption goals across all teams: role-based enablement sessions, golden path demonstrations, governance ownership definitions, and a metrics framework for tracking adoption and impact over the following 3 to 6 months.
41 production-ready deliverables across seven interconnected frameworks
Brownfield Analysis Engine. A four-pass, AI-assisted process for reverse-engineering undocumented legacy codebases into trustworthy architecture documentation. The engine scans repository structure and dependencies, infers architectural decisions with confidence scoring, validates findings with senior engineers through structured human-in-the-loop review, and produces final C4 models generated from actual code. For Vanco's roughly 70% legacy codebase, this turns the single biggest scaling bottleneck, tribal knowledge dependency, into navigable documentation any developer can use on day one.
30 Playbooks. 18 role-specific playbooks (from Architect to UX Designer) and 12 on-demand task playbooks covering common workflows like generating architecture diagrams, enhancing acceptance criteria with Gherkin format, generating tests, and performing delta analysis. Each is calibrated to Vanco's specific stack, conventions, and organizational structure.
Context Packs. Repository-specific knowledge bundles that feed AI tools with Vanco's actual patterns and constraints, so generated code compiles and follows Vanco standards out of the box.
SDLC Transformation Blueprint. A comprehensive AS-IS to TO-BE guide covering all six SDLC stages, with explicit Human/AI boundary definitions at three tiers.
Supporting Assets. Story and bug templates with Definition of Ready/Done standards, Cursor IDE setup guide, cursor-rules repository, and reference implementation repositories demonstrating the complete golden path.
Every SDLC stage in the blueprint defines who decides, who drafts, and what runs under governed automation.
The transformation, before and after
Eight parts of Vanco's delivery system changed shape. Toggle between where each one started and where the VancoSDLC Framework puts it.
Requirements6 bullet points per epic, under 15% acceptance-criteria field usage
Architecture docsDrawn from memory in LucidChart, frequently outdated
Legacy knowledgeTribal, concentrated in senior engineers
AI usageUngoverned Copilot accelerating code without standards
TestingAbout 20% automated, QA discovering requirements
OnboardingMonths to become productive on legacy codebases
Rework10 to 15% of active work was reversions
ProcessStage-gated, approval-driven despite Scrum ceremonies
RequirementsStructured stories with Gherkin ACs, edge cases, integration flags, and test scenarios
Architecture docsC4 models generated from actual code via the Brownfield Analysis engine
Legacy knowledgeNavigable architecture documentation any developer can use on day one
AI usageGoverned AI workflows with Context Packs ensuring output matches Vanco patterns
TestingAI-generated unit, integration, and E2E tests from structured ACs; QA validating, not discovering
OnboardingBrownfield docs, Context Packs, and role playbooks compress ramp-up
ReworkUpstream quality enforcement and DoR/DoD compliance reduce rework at source
ProcessDefined SDLC with clear Human/AI boundaries and governed automation at every stage
The engagement in numbers
team members trained across 8 roles
production-ready deliverables across seven frameworks
playbooks: 18 role-specific, 12 on-demand
SDLC stages transformed, AS-IS to TO-BE
A four-pass Brownfield Analysis engine that produces C4 models from actual code
Three explicit Human/AI boundary tiers at every SDLC stage: Human Decision, AI Assist, AI Automate
A metrics framework for tracking adoption and impact over the following 3 to 6 months
How GenDD powered the engagement itself
This engagement was itself executed using GenDD principles. The methodology informed every phase of the work, so the deliverables embody the same loop they encode.
-
01
Workflow-first mindset
Before selecting tools or writing templates, HatchWorks mapped how work should flow at each SDLC stage, defining inputs, quality gates, outputs, and handoffs. The tools serve the workflow, not the other way around.
-
02
AI-assisted acceleration
Playbooks, templates, architecture documentation, and test scaffolds were all generated with AI assistance and then validated, refined, and hardened through human review. The deliverables embody the same generate, validate, review, harden loop they encode.
-
03
Contract-first thinking
API specifications and interface contracts were defined before implementation, enabling parallel development and reducing integration churn. This same principle was encoded into the story templates and architecture scaffolds.
-
04
Pattern capture
Every solution discovered during the engagement was encoded into a reusable scaffold so every team benefits from prior work. The system compounds: each project makes the next one faster.
The engagement didn't just produce GenDD artifacts. It was itself executed using GenDD principles.
In their words
“We wanted repeatability and oversight. HatchWorks AI helped us build an approach that scales responsibly across teams. This work gave us a safer, more standardized way to apply GenAI in delivery.”
Steven JoosChief Product Officer
About the work
About Vanco
Vanco is a leading provider of secure digital payments, online giving, and administrative software for faith communities, schools, and nonprofits. Vanco helps organizations accept and manage payments, streamline everyday tasks, and deepen participation with reliable tools and practical support. The company's faith business will now operate as ACS Technologies, offering trusted church management platforms such as Realm and MinistryPlatform. Vanco also serves K-12 districts with education payments solutions, including RevTrak and SmartCare. Across every offering, Vanco's purpose is simple: help teams spend less time on systems and more time with people.
HatchWorks AI turns AI into ROI by automating key business processes, transforming data, deploying intelligent agents, and shipping AI-powered products that deliver measurable results. Its proprietary Generative-Driven Development methodology is a repeatable path from idea to production, blending AI, agents, and engineering to ship faster with less risk.
Ready to turn ad-hoc AI into an operating model?
See how HatchWorks AI takes engineering organizations from isolated experiments to governed, repeatable AI-augmented delivery.