-
-
Landing page
Inspiration
AI agents are becoming increasingly capable, but capability alone does not create trust.
Organizations are beginning to rely on AI for software development, operations, analysis, and decision support, yet most agent systems still operate as black boxes. They can generate outputs, but they often lack accountability, verification, and visibility into how work gets done.
We became interested in a different problem. Instead of asking how to make AI smarter, we asked:
How do we make AI execution trustworthy?
That question inspired Brewing. We believe the future will involve humans and autonomous systems working together, but that future requires infrastructure for coordination, verification, observability, and accountability.
For this hackathon, we demonstrate that vision through a real-world software engineering workflow.
What it does
Brewing is trust infrastructure for AI execution.
A user submits an objective such as reviewing a repository for production readiness. Brewing coordinates specialized AI agents to complete the objective. The platform creates an execution plan, retrieves repository information through GitLab MCP, assigns tasks to specialist agents, validates results through a verification layer, and provides visibility into the entire process through observability tooling.
Instead of stopping at AI-generated outputs, Brewing focuses on trustworthy execution.
The result is a system that helps humans confidently collaborate with AI while maintaining oversight and accountability.
How we built it
We extended Brewing's existing infrastructure for objective management, agent coordination, governance, verification, and workflow orchestration.
Gemini powers planning, reasoning, and task coordination.
GitLab MCP enables real-world repository access and software engineering workflows.
Arize provides execution visibility, tracing, and observability into agent behavior.
The system follows a structured workflow:
Objective → Planning → Execution → Verification → Outcome
Specialized agents perform different responsibilities, while a verification layer validates outputs before they are accepted.
Challenges we ran into
One of our biggest challenges was avoiding the temptation to build another AI chatbot.
Many AI systems generate answers, but real-world workflows require verification, observability, auditability, and mechanisms for handling uncertainty. Designing a system that remains useful when things go wrong proved significantly harder than generating outputs.
Another challenge was coordinating multiple agents while maintaining clear visibility into their actions. We wanted every major decision and execution step to be understandable, traceable, and reviewable.
Finally, we had to balance automation with human oversight. Our goal was not to replace human judgment, but to strengthen it.
Accomplishments that we're proud of
- Built a working objective-driven multi-agent workflow.
- Extended an existing trust infrastructure platform instead of creating a simple demo agent.
- Integrated real-world repository workflows through GitLab MCP.
- Added verification and accountability mechanisms to agent execution.
- Enabled observability into agent actions and execution outcomes.
- Demonstrated how AI systems can move beyond generation toward trustworthy execution.
Most importantly, we built a system that focuses on trust rather than simply intelligence.
What we learned
The biggest challenge in agentic systems is not intelligence.
It is trust.
As AI becomes more capable, organizations need more than powerful models. They need systems that provide verification, accountability, visibility, and confidence in outcomes.
We learned that production-ready AI requires much more than prompting. It requires governance, observability, correction mechanisms, and thoughtful human oversight.
Those lessons shaped every design decision in Brewing.
What's next for Brewing
Repository auditing is only the beginning.
The same architecture can be applied to compliance reviews, operational workflows, software delivery pipelines, enterprise processes, and other high-trust environments where organizations need confidence in AI-assisted execution.
Our long-term vision is to make Brewing the trust layer that enables humans and autonomous systems to coordinate work safely, transparently, and effectively.
As AI becomes increasingly integrated into economic and operational systems, we believe trustworthy execution will become one of the most important infrastructure challenges of the decade.
Log in or sign up for Devpost to join the conversation.