Philosophy and Core Principles
WebbyLab strictly adheres to the principle of "Human in the Loop". We utilize generative AI tools, frontier LLMs, and AI agents exclusively as Co-pilots — accelerating routine operations, facilitating refactoring, automating repetitive workflows, and expanding our engineers' capacity for high-value problem solving.
AI accelerates our engineering work — but every commit is reviewed, understood, and signed off by a senior engineer. You always receive production-ready software validated by an expert, not an unchecked AI output. Our engineers take full ownership of every line that ships: quality, security, and correctness are non-negotiable.
Enterprise AI Development Stack & Infrastructure
Authorized Access
Access to all AI development tools is granted exclusively via Google Workspace Enterprise SSO. Personal accounts for corporate tasks are categorically prohibited — your intellectual property stays protected at every level.
Approved Models & Tools
The company utilizes the Antigravity platform, powered by state-of-the-art frontier LLMs: Google Gemini (Vertex AI) and Anthropic Claude (Opus and Sonnet). Model selection is matched to task complexity, balancing output quality and cost efficiency.
Corporate MCP & AI Agent Skills Framework
We centralize our 15 years of engineering expertise into corporate Model Context Protocol (MCP) servers, powering standardized AI agent workflows and Skills shared across every team. This institutional knowledge — not generic AI output — is what drives consistent delivery quality.
- System Architecture scaffolding & AI-assisted design patterns
- Automated Code Review & AI-powered Refactoring
- Domain-Driven Design (DDD) workflows
- Clean Architecture enforcement
- Atomic Design & React component best practices
Enterprise AI Security Controls (Antigravity)
AI assistants within the IDE operate only under an API Key (Bring Your Own Key — BYOK) model. SaaS subscriptions where your code is processed on external servers are strictly forbidden. Your source code never leaves your environment.
- Set
"Codebase Indexing"to Local - Completely disable "Data Collection / Training"
AI Integration Across the Full SDLC
Our Corporate RAG (Retrieval-Augmented Generation) system, powered by LLMs, analyzes historical task estimations against actual completion times across projects. As AI-assisted development continuously increases engineering velocity, this data allows us to deliver progressively more accurate timelines and lower feature development costs for our clients.
Generative AI helps surface ambiguities in business requirements and acceptance criteria before a single line of code is written — reducing costly rework, misaligned deliverables, and scope creep downstream. Better inputs mean better software, faster.
Our RAG system surfaces battle-tested architectural solutions from across our entire project portfolio — so your product benefits from non-trivial patterns already proven in production. Cross-project sharing follows the Client AI Consent Questionnaire; disabled on opt-out.
Teams combine MCPs and specialized AI Skills to generate UI kits directly into the application framework, dramatically shrinking the design-to-code cycle. The Antigravity Chrome extension visually verifies component fidelity against Figma specs — so what you designed is exactly what gets built.
Every repository contains an active .Antigravityrules file defining the AI co-pilot's role, tech stack, coding standards, and hard constraints (e.g., "no 'any' in TypeScript"). This ensures AI-generated code aligns with your architecture — not generic boilerplate. The Tech Lead owns and maintains this file.
Routine tasks (standard forms, CRUD APIs) are AI-generated via battle-tested Workflows, freeing engineers for complex work. Where custom business logic exceeds AI capabilities, engineers build a manual Proof of Concept (PoC) first — then guide the AI to integrate it correctly.
AI assists threat modeling under STRIDE and DREAD methodologies as a supplementary analysis tool, helping teams identify attack vectors earlier in the development cycle. Final security decisions rest with human architects. Sensitive client data (PII, credentials) never enters any AI model — enforced at the process level, not just policy.
QA and engineering teams apply a Test-Driven Development (TDD) approach powered by AI. Acceptance Criteria from the issue tracker are fed directly into AI/MCP tools, accelerating generation of test cases and automated tests (Mocha, Chai, PHPUnit). Result: broader coverage in less time — your product ships with confidence, not fingers crossed.
Following static analysis and automated testing, code undergoes an additional AI-driven review via a custom, self-hosted GitLab MCP on our internal infrastructure. AI catches issues before human review — engineers focus their attention where it matters most.
All AI-generated code still requires mandatory human Code Review. No AI output reaches production without expert sign-off — giving you a fully auditable development trail and assurance that every release has been validated by a senior engineer.
A dedicated MCP handles debugging and AI-assisted analysis of logs from Amazon CloudWatch and other observability vendors. AI surfaces anomalies and root causes faster than manual trawling — directly reducing your product's mean time to resolution (MTTR).
Before any logs reach an AI model, engineers sanitize them — stripping all secrets, credentials, and personal data. Your users' data stays fully protected even during active incident response. Privacy compliance is built into the debugging workflow.
During Sprint Retrospectives, the team tracks AI integration effectiveness — including tokens consumed per feature — enabling continuous optimization. As tooling matures, this translates into compounding velocity improvements and ongoing cost reductions for your product.
Token expenditures are tied to each project's Cost Center. Where AI tokens power core Product AI features, these are treated as recurring infrastructure costs and billed directly via Direct Billing or Reimbursement as agreed in the Client AI Questionnaire. No hidden AI costs.