AI gets thrown at implementation too quickly. Without clear requirements. Without architectural decisions. Without a thought-through plan.
The result is what many teams experience: inconsistent code that doesn’t survive review, hard-to-maintain solutions, and a team questioning whether AI actually helps or just produces chaos faster.
At innFactory and innFactory AI, we have deliberately chosen a different path. We call it Spec-Driven Development – and it is fundamentally changing how we build software.
The Real Problem: Structure Comes Before Code
Throwing AI directly at implementation skips exactly the steps that make good software development: thorough requirements analysis, a clear architecture, and a well-thought-out implementation plan.
This isn’t a new problem. It’s the old problem – just made visible at a new speed.
AI amplifies what’s already there. Teams with a solid engineering foundation get significantly better results with AI. Teams that start without structure get bad results faster with AI.
The decisive shift isn’t the tool. It’s the thinking that comes before it.
Spec-Driven Development: Think First, Build Second
Our approach follows a clear sequence:
Specify → Validate → Implement → Review
Before a single line of code is written, we develop three core documents together with the AI agent:
- Product Requirements – What should be built and why?
- Solution Design – How is it implemented technically?
- Implementation Plan – What phases, tasks, and dependencies?
Only once these documents exist and have passed quality gates do we move to implementation. The advantage: AI doesn’t work blind, but follows clearly defined guardrails. The result is more consistent, maintainable, and review-ready code – without sacrificing creativity.
Between phases, automated quality gates check for Completeness, Consistency, and Correctness before the next phase is unlocked.
The Tool Triad at innFactory
Spec-Driven Development requires the right tooling. At innFactory, we’ve chosen three tools that together form a complete AI development stack.
OpenCode – Agent-Driven Development Flow
OpenCode is the heart of our development flow. This open-source AI coding tool goes far beyond classic code completion: it performs complete multi-file refactorings, understands cross-project relationships, and coordinates specialized agents for different tasks.
What makes OpenCode special: it’s model-agnostic. With support for over 75 AI providers, we can use the optimal model for each task – without vendor lock-in, with full control over data and infrastructure.
OpenCode also brings a powerful skill and agent system that we’ve heavily customized at innFactory to match our own templates, coding standards, and architectural principles. More on that below.
GitHub Copilot via Azure – AI in Every Developer’s Daily Work
As a Microsoft Partner, we provide our developers with GitHub Copilot with access to all available models through Azure. This enables AI support directly in the IDE – for code completion, tests, documentation, and code review.
GitHub Copilot is the daily companion: always available, seamlessly integrated into existing workflows. We train our developers specifically to use Copilot not just as autocomplete, but as a real pair programmer – with clear prompts, critical review of results, and awareness of the tool’s limitations.
As a Microsoft Partner, we can also introduce, configure, and accompany GitHub Copilot for other companies with appropriate training formats.
CompanyGPT – GDPR-Compliant AI for the Entire Company
CompanyGPT is our company-wide AI platform – and much more than a ChatGPT replacement.
CompanyGPT is a privacy-compliant, Europe-hosted AI platform based on LibreChat that provides access to all leading language models: GPT-4o, Claude, Gemini, Llama, and many more – in a centralized, controllable environment. No data goes to OpenAI or other US providers, no company data is used for training external models.
For developers, CompanyGPT is the tool for everything beyond code: architecture discussions, technical documentation, explaining complex concepts, communication, and analysis. We also use CompanyGPT for internal knowledge management and as the foundation for company-specific assistants.
As operators of CompanyGPT, we can also provide the tool to other companies, set it up, and accompany them with training.
Specialized Agents for Every Role
A central element of our setup is the agent framework, which we’ve tailored to our own structures, coding conventions, and architectural principles.
Different specialized agents support different roles in the development team:
- The Analyst – Requirements analysis, market and user research
- The Architect – System design, security and robustness reviews
- The Developer – Feature implementation, performance optimization
- The DevOps – Infrastructure, CI/CD, production monitoring
- The Designer – UX research, interaction design, visual design
- The Tester – Test strategy, coverage analysis, edge cases
- The Scala Developer – Specialized in our Scala/Play projects with innFactory-specific patterns
Each agent knows its tasks, its boundaries, and our innFactory-specific rules. This ensures the entire team follows the same development flow – regardless of the project or which developer is working on it.
A constitution layer ensures that project-wide governance rules are respected in every step of the workflow.
Bringing Everyone Along: Training Is Not an Afterthought
Technology alone doesn’t transform anything. That’s why we invest heavily in various training formats – with the goal of truly bringing all developers along.
Our enablement approach includes:
Internal talks and presentations – Regular sessions where we share and discuss new tools, approaches, and learnings.
Video trainings – Our own training videos that show and demonstrate best practices and concrete workflows. Not theoretical knowledge, but real workflows from daily work.
Developer Experience Day – Our next DX Day in March offers full-day, deep sessions on AI-driven development, tools, and workflows.
Pairing and code reviews – Experienced developers work directly with the team, demonstrating patterns and providing feedback on AI-generated code.
The goal: no developer gets left behind. AI-driven development is not a privilege for early adopters – it’s the new normal.
What This Means for the Future of Knowledge Work
What we’re experiencing in software development right now is only the beginning.
During my trip to San Francisco as part of the Bavarian Ministry of Economic Affairs’ delegation to the Bay Area, one of the strongest signals was exactly this: The transformation that software development is currently undergoing will soon extend to all areas of knowledge work.
The work of analysts, project managers, architects, consultants, lawyers – in short, all classic white-collar workers – will fundamentally change. Not because jobs will disappear, but because the nature of work will shift: away from manual execution, toward structured thinking, quality assurance, and strategic direction.
At innFactory AI, we are convinced: those who actively shape this transformation – with structure, the right tools, and a focus on enablement – will have a decisive advantage.
Conclusion: Structure Makes the Difference
Spec-Driven Development is not a new process. It’s the consistent application of good engineering principles to a world with powerful AI assistants.
AI doesn’t work better because the model is larger. It works better because we give it clear structure: requirements, architecture, plan – and then implementation.
With OpenCode, GitHub Copilot via Azure, and CompanyGPT, we have a tool stack that enables exactly that. With our agent framework and training formats, we ensure the entire innFactory team benefits from this approach.
If you want to learn more about Spec-Driven Development, our tools, or transforming your own development processes – get in touch.
Tobias Jonas is Co-CEO of innFactory AI Consulting GmbH and innFactory GmbH. He guides companies through strategic and technical AI transformation.