By: Darren Mills
Most organizations experimenting with AI are still operating in an ad hoc model: individual teams adopting disconnected tools, workflows, and prompting strategies. That can create pockets of productivity, but it rarely creates the kind of scalable engineering model needed to deliver consistent, governed, and repeatable outcomes across the software delivery lifecycle.
CleanSlate Technology Group approached AI adoption differently. The goal was not to hand the team another tool and hope for the best. The goal was to build a platform engineering and delivery governance model for AI-assisted work across architecture, development, testing, project delivery, and technical documentation.
The Problem: Ten Engineers, Ten Different Tools
CleanSlate Technology Group, an AWS Premier Tier Partner based in Indianapolis, surveyed its engineering organization and found a fragmented AI ecosystem. Nearly every engineer was using a different platform, including ChatGPT, GitHub Copilot, Claude, and other emerging tools.
The result was predictable:
- Inconsistent delivery artifacts and engineering outputs
- Limited governance and oversight
- No repeatable operating model for AI-assisted engineering practices
- Increased validation overhead and delivery risk
The organization recognized that AI adoption required more than selecting a preferred toolset. CleanSlate needed three things:
- A standardized AI engineering platform across delivery teams
- Governance controls including policy enforcement, validation workflows, architectural standards, and operational safeguards
- Reusable engineering accelerators, architectural patterns, and spec-driven delivery frameworks
The goal was not simply to find a better AI tool. It was to establish a governed, AI-enabled delivery operating model that could scale across projects, roles, and client engagements.
Enter Kiro: Beyond AI-Assisted Development
As an AWS Premier Tier Partner, CleanSlate evaluated Kiro as a specification-driven engineering platform capable of orchestrating requirements management, architecture design, implementation workflows, and AI-assisted delivery practices.
Built and managed by AWS, Kiro enables engineering teams to standardize AI-assisted development through reusable specifications, workflow orchestration, and structured delivery patterns. Its specification-driven approach transforms structured requirements into architecture artifacts, implementation workflows, validation patterns, and deployable code.
But what attracted CleanSlate wasn’t just the coding capability. It was the chance to make AI part of how delivery work actually gets done.
“Kiro allows you to synthesize documents, generate reports, build presentations, combine transcripts, notes, and structured data into delivery artifacts. It becomes an operational platform for how engineering teams deliver work, not just how they generate code.”
— Stephen Henderson, Technical Manager, CleanSlate
That distinction matters. Kiro is not a productivity shortcut. For CleanSlate, it became the backbone for how teams generate architecture, reports, client deliverables, and engineering work products across the business, not just how developers generate code.
“We’ve built a governed operating model for AI-assisted delivery that spans development, project management, assessments, UI/UX, and testing. This extends beyond writing code into the entire software delivery lifecycle.”
— Darren Mills, Chief Technologist, CleanSlate
Framework First, Tool Second
One of CleanSlate’s key lessons was simple: the framework has to come before the tool.
“Other companies hand their teams a tool and say, ‘go use it.’ We took the opposite approach: define the operating model first, then embed the tooling into governed workflows with standardized delivery patterns.” — Mills
The initiative began with a focused pilot group of approximately 8-10 team members working across several active projects. Before adopting Kiro for production delivery, the team invested in:
- Reusable steering frameworks
- Domain-specific skill libraries
- Engineering and architecture standards
- Validation and review workflows
- Governance boundaries and operational safeguards
This foundation established consistent expectations for delivery quality, traceability, and engineering outcomes before AI-assisted workflows were introduced at scale. That is what made adoption feel less like experimentation and more like an operating model.
The Solution: A Five-Layer Governed Engineering Platform
CleanSlate did not simply adopt Kiro as a coding assistant. The organization built a five-layer operating model around it, turning Kiro into the foundation for how different roles plan work, apply standards, validate outputs, and deliver client-ready artifacts.
Layer 1: Agents: Who Does the Work
CleanSlate implemented domain-specific AI personas with scoped permissions and operational boundaries. Engineers use infrastructure-aware agents aligned to AWS CDK and governance controls. Project managers leverage agents optimized for specification-to-sprint conversion. Sales teams utilize agents designed for branded SOW generation and proposal consistency. Each agent operates within clearly defined constraints to maintain governance and delivery consistency.
Layer 2: Steering: What Rules the AI Follows
Steering files became the mechanism for embedding governance directly into AI-assisted workflows. Architectural standards, naming conventions, security requirements, diagram formatting rules, and delivery expectations are automatically loaded based on project context and file type.
Rather than relying on individual prompting discipline, CleanSlate standardized governance through reusable markdown-driven delivery controls. Steering standards are version-controlled within a centralized repository, ensuring engineering consistency across all projects and delivery teams.
Layer 3: Skills: What Expertise the AI Has
CleanSlate codified internal consulting methodologies into reusable Kiro skills. These skills encapsulate enterprise architecture frameworks, application modernization approaches, assessment methodologies, enterprise planning models, and branded reporting standards.
Skills activate dynamically based on workflow context, enabling teams to apply repeatable consulting methodologies without manually loading supporting documentation or guidance.
Layer 4: Hooks: What Guardrails Are Enforced
Hooks provide automated governance enforcement within the delivery lifecycle. CleanSlate implemented controls that:
- Block destructive infrastructure commands without explicit approval
- Scan modified files for secrets and sensitive data
- Require planning validation before authentication or security modifications
- Inject Git branch and AWS account context into sessions automatically
These controls operate as event-driven governance mechanisms rather than manual review checkpoints.
Layer 5: Specs: How Work Is Planned
Kiro’s specification-driven workflow allows natural language requirements to evolve into structured requirements documentation, architecture decisions, sequenced implementation tasks, and sprint-ready delivery plans.
CleanSlate uses this approach for feature development, infrastructure planning, modernization initiatives, and project execution workflows, integrating specifications directly into Jira and Azure DevOps delivery pipelines.
The Governance Breakthrough: Context, Not Prompting
The technical breakthrough was not better prompting. It was better context. Prompting felt unnatural until engineers had the right structure around them. Once steering files, skills, hooks, and pre-loaded agents were in place, the learning curve dropped quickly.
CleanSlate built a custom MCP (Model Context Protocol) server that reads the current working directory, matches it to the appropriate project, and serves merged steering files live. Every engineer, in every session, gets the same standards automatically.
That changed the conversation. Engineers stopped asking, “How do I prompt this?” and started asking, “What should I build next?” Governance moved out of static documentation and into version-controlled delivery workflows that load automatically where the work happens.
Beyond Code: How Each Role Uses the Platform
The biggest shift is that Kiro is not limited to engineering tasks. The same governed model now supports the way CleanSlate teams design, plan, sell, and deliver work.
- Sales teams: A sales lead can generate a branded SOW or proposal draft that reflects CleanSlate’s voice, delivery model, assumptions, and accelerator language without starting from a blank page.
- Project managers: A project manager can convert approved specifications into sprint-ready backlog items, migration waves, and delivery plans that integrate with Jira or Azure DevOps.
- Architects: An architect can turn meeting transcripts into Mermaid diagrams and AWS-aligned architecture artifacts, with steering files applying the same design conventions and diagram standards across projects.
- Engineers: An engineer can start with a feature specification, generate implementation patterns, and have governance hooks check context, secrets, and risky infrastructure actions before the work moves forward.
- Delivery teams: A delivery team can run modernization assessments, IT planning exercises, and executive reporting through reusable frameworks that keep findings consistent from one engagement to the next.
The Results: When the Framework Clicks
Once the foundation was in place, the results started to show up in the work itself.
Work that previously suffered from inconsistent outputs, hallucinations, and repeated validation cycles became more consistent and repeatable. Teams reduced rework, accelerated delivery execution, and began identifying flaws and architectural gaps earlier in the lifecycle.
That is where the platform started to feel different. It was not just faster output. It was more reliable output with clearer review points and stronger delivery controls.
Henderson described a recent example after a client meeting:
“I was able to generate architecture artifacts, produce technical diagrams, and assemble executive-ready client deliverables, all within about an hour. But it’s not magic. You need the right inputs, structure, and governance framework behind it.”
That caveat is important. The productivity gains aren’t accidental. The acceleration came from the governance framework established before platform adoption, not from the AI tooling alone. The acceleration came from combining structured governance, reusable frameworks, and disciplined delivery practices with AI-assisted workflows.
The Bigger Shift: Changing What Engineers Do
The gains go beyond speed. CleanSlate’s engineers are changing how they work, not just how fast they work.
At the leadership level, Mills sees the role of engineers evolving from primarily writing code to designing specifications, governance models, and validation frameworks that guide AI-assisted implementation.
“We’re moving toward writing technical specifications, allowing AI to generate implementation patterns, and then validating and governing the outputs. That fundamentally changes how we structure software delivery.” — Mills
This operating model extends beyond internal engineering practices. CleanSlate now incorporates the same governed AI workflows, reusable accelerators, and validation frameworks into client-facing engagements, including Application Modernization assessments and AWS Well-Architected Framework Reviews.
Rather than simply advising clients on AI adoption, CleanSlate operates an AI-enabled delivery model and brings those practices directly into customer engagements.
Lessons for Organizations Considering AI Adoption
CleanSlate has been candid about what they’ve learned, especially because the lessons are practical rather than theoretical.
Adoption matters more than the tool
- Prioritize organizational readiness before deploying AI platforms broadly
- Unstructured adoption creates delivery inconsistency, governance gaps, and operational risk
- Long-term value comes from shared standards, repeatable workflows, and disciplined practices
Establish Governance Before Scaling
- Define steering files, standards, and delivery expectations early
- Build validation workflows and review processes before introducing AI into production delivery
- Strong governance foundations create downstream reliability and repeatability
AI Value Depends on Workflow Design
- Structured inputs and intentional workflows significantly improve outcomes
- AI can support architecture, documentation, reporting, planning, and engineering workflows when properly governed
- Engineering maturity increasingly depends on how effectively organizations guide and validate AI-assisted delivery
To support this model, CleanSlate implemented governance standards around prompt management, reusable delivery artifacts, human review gates, and validation workflows to ensure consistency, traceability, and alignment with enterprise delivery and security requirements.
Hear It Directly from the Team
CleanSlate Technical Manager Stephen Henderson, accompanied by CTO Darren Mills, will be speaking at the Indy AWS Meet-Up at CleanSlate’s corporate headquarters on April 28th, 2026.
They will discuss how their team built its Kiro framework, lessons learned during adoption, and how governed AI delivery is reshaping engineering operations.
Location: 645 W. Carmel Drive, Suite 140, Carmel, Indiana 46032
Date: April 28, 2026
Registration is open: https://www.meetup.com/indyaws/events/314430843/
Attendees will have the opportunity to see these workflows in practice, engage directly with the engineering team, and learn how CleanSlate helps organizations move from isolated AI experimentation to operationalized, enterprise-scale delivery.