Scaling Quality Without Slowing Engineering
MongoDB’s Atlas API Platform team acts as a force multiplier – enabling hundreds of engineers to build consistent, performant APIs across MongoDB’s cloud ecosystem.
Maintaining quality meant balancing two pressures:
- Upholding strict API design and performance standards
- Avoiding process overheads that slows delivery
Traditional automation (like linting or static validation) couldn’t catch nuanced architectural or performance issues. The team needed a system that could reason across large codebases and adapt to team API guidelines.
“Most assistants focus on code generation, but that was never our bottleneck,” said Wojciech Trocki, Engineer working on MongoDB Atlas API Platform team. “The real value comes from reasoning — understanding architecture, testing assumptions, and enforcing standards by automation”
From Prompting to Specialized Agents
To evaluate AI tools, MongoDB formed a cross-functional working group of engineers who tested AI IDE assistants across projects. Their criteria focused on reasoning, scalability, and integration.
“My team didn’t want a glorified autocomplete. We wanted an expert partner — something that could reason about a large codebase and help us make better technical decisions,” Wojtek explained.
Augment stood out for three reasons:
- Deep codebase reasoning – reliable search and contextual understanding across large codebases.
- Multi-step execution – chaining multiple commands and reasoning steps to solve complex problems.
- Flexible platform – IDE integration, CLI, and MCP connections for internal systems like CI/CD.
“The CLI changed everything,” Wojtek added. “Being able to call agents directly from my existing workflows — CI/CD, validation, architecture reviews — that’s what made it real for me and the engineers in my team”
Building a Framework for AI Maturity
Wojciech and his team build upon an a four-level maturity model for AI adoption:
- Prompting – establish structured queries and use cases.
- Contextualization (MCPs) – connect Augment to internal systems for richer context.
- Guidelines & Guardrails – encode architectural rules into Augment projects, both auto and manual.
- Specialized Agents – build task-specific agents via Augment CLI.
“We realized prompt engineering was just step one,” Wojtek said. “The real maturity comes when you start encoding your team’s principles — guidelines, rules, context — and let Augment Code act on them.”
Results
From Insight to Automation
1. Architecture Validation Agent Automates design reviews by reasoning over technical specs, internal guidelines, and prior implementations — surfacing risks and large combinations of alternative solutions before review.
2. Performance Analysis Agent Processes CI/CD logs (often hundreds of MBs) to identify anomalies and regression causes overnight, accelerating debugging cycles.
3. Dynamic Guidelines Engine Transforms MongoDB’s API Standards site into a living MCP source. Agents query rules dynamically rather than relying on static markdown files, improving accuracy and maintainability.
“You don’t need to replace your processes to use AI,” Wojciech said. “Augment fits into existing process bridging knowledge gaps — it just removes friction.”
Lessons Learned
- Start simple: Begin with a few high-value use cases.
- Codify culture: Augment guidelines and guardrails are more powerful than prompts.
- Invest in plans: Effective agents using Augment CLI start with structured, human-level plans.
- Empower champions: Volunteer groups accelerate learning and adoption.
“The biggest productivity gain wasn’t in writing code faster — it was in reducing the back-and-forth,” said Wojtek. “The agent gives you feedback before the pull request even exists.”
What’s Next
MongoDB’s next focus is scaling these patterns across engineering — promoting reusable Augment CLI workflows and expanding AI-assisted PR validation and review.
By combining Augment’s reasoning capabilities with MongoDB’s engineering rigor, the team is turning AI from an assistant into infrastructure.
“Our rule was simple,” Wojciech summarized. “AI should enhance our culture of quality, never compromise it.”