Week 6 of 8 · Build AI
Architecture, integrations and security
Design the codebase before connecting it: the principles, then integrations into the systems a business runs on, the config-first monorepo that holds them, and the security that protects them.
What the week builds toward
WorkBot redesigned on the architecture principles and moved into a config-first monorepo; connected to a CRM, a ticketing system, a calendar and GitHub over a unified, rule-applied data layer; configuration per customer with a rule language and a generator; roles, row-level security, personal-data handling and managed secrets.
The 5 days
- Day 26
Architecture principles: what to unlearn
5 architectures to drop and the config-first, lightweight design to adopt, applied to WorkBot.
You ship: An architecture audit, 2 migrations measured before and after, and a decision record.
- Day 27
Integration platforms: Composio, Nango and MCP
Managed auth, syncs and MCP as 3 routes into a customer's SaaS, and the tool surface on top.
You ship: The agent connected to a CRM, a ticketing system and GitHub through 3 routes, wrapped in product-level tools.
- Day 28
Integration techniques: sync, unified schemas, rules and caching
Incremental sync, one schema across providers, deterministic rules, and per-field caching.
You ship: A unified, rule-applied, cached data layer under the integrations, with reconciliation and an audit trail.
- Day 29
Config-first design, grammars and meta-compilers
Behavior as configuration, a BNF grammar and parser for a rule language, a generator, and the monorepo that holds them.
You ship: WorkBot in one config-first monorepo, with a second tenant from configuration, a rule language with 20 tests, and an entity generator.
- Day 30
Agent security: RBAC, keys and data separation
Authorization enforced below the model, tenant separation in every store, key and secret handling, and injection defenses tested against an eval set.
You ship: WorkBot integrated, config-first and hardened, with an injection eval set it fails.