Build Atlas
System design has shared words like database, queue and cache. Building with AI needs the same kind of words for everything else, so you can name what you need and tell an agent to build it.
Each entry explains what a component is, when you need it and the popular choices, and ends with an instruction you can hand to your agent.
Product
Deciding what to build and learning whether it works for people.
The application
Components inside the software you ship.
- App frameworkThe foundation that handles routing, rendering and server code for a web or mobile app.
- AuthenticationSigns users in, keeps them signed in, and controls what they can access.
- Infrastructure from codeDeclares databases, queues and services in code, so the same setup runs locally and in the cloud.
- Runtime schema validationChecks that data coming into your app has the shape you expect, at the moment it arrives.
- State machine / statechartWrites complex logic as explicit states and transitions instead of scattered flags.
- Telemetry (traces, metrics, logs)Records what your running system is doing, so people and agents can see why something went wrong.
- Typed errors and effectsMakes the ways a function can fail part of its type, so callers have to handle them.
- UI component workshopA place to build and check UI components in every state, outside the full app.
Testing
Checks that catch what the author, human or agent, didn’t think of.
- Accessibility checkerAutomatically finds accessibility problems like missing labels, low contrast and broken keyboard access.
- Component and end-to-end testsTests the UI in a real browser, from single components up to full user journeys.
- FuzzerThrows huge numbers of generated, often malformed inputs at your code to find crashes.
- Property-based testingTests a rule against hundreds of generated inputs instead of a few you picked by hand.
- Visual regression testingCatches unintended visual changes by comparing screenshots against a saved baseline.
AI in the system
Components for products that use LLMs and agents.
- Agent and MCP security scannerChecks AI agents, MCP servers and skills for prompt injection, tool poisoning and leaked secrets.
- Document parsing for RAGTurns PDFs, Word files and slides into clean text and structure that a model can use.
- LLM evalsAutomated tests for AI features, which score outputs against examples and criteria.
- LLM SDKOne API for calling different AI models, with tool calling and structured output built in.
- LLM tracing and prompt managementRecords every model call, with its prompt, output, cost and latency, so you can debug AI features.
- Voice agent pipelineConnects speech recognition, a language model and speech synthesis into a real-time conversation.
Building with agents
The repo, workflow and guardrails that keep coding agents productive and honest.
- Agent instructions file (AGENTS.md)A file in your repo that tells coding agents how the project works and how to check their work.
- Agent sandboxLimits what an agent can touch on your machine and network, so it can work unattended safely.
- Agent skillsPackaged procedures and know-how that an agent loads only when a task needs them.
- Agent task trackerA task list agents can read and update themselves, so long-running work doesn't get lost between sessions.
- AI code reviewA separate agent that reviews changes against the spec and your standards before a human does.
- Custom lint rules and architecture checksTurns your team's rules into checks that fail the build, with error messages that tell the agent how to fix it.
- Duplication and dead-code scannersFinds copy-pasted logic and unused code, giving cleanup work a concrete list.
- Git hooks and CI gatesRuns your checks automatically before commits and merges, so broken work can't slip through.
- Specs and execution plansWritten requirements and a living plan, so agents build the right thing and long work survives across sessions.