# How I Like to Build — Engineering Manifesto (llms.txt)

> Canonical URL: https://tobilobaodejinmi.com/manifesto
> Author: Tobiloba Odejinmi (https://tobilobaodejinmi.com/about)
> Context: Three core engineering rules formed across 12+ years of production systems.

---

## The Three Rules

### Rule 01: Make it easy to plug in
> "If another team can't use your API before lunch, there's more work to do."

- **The Principle**: Software that requires weeks of coordination, ambiguous schemas, or undocumented side-effects creates friction. Modern architectures should expose strict, typed OpenAPI contracts, clean webhooks, and idempotent interfaces that an engineer can understand, test, and integrate in a single morning.
- **Application in AI Employees**: An autonomous agent should never require adopting a completely new platform or vendor lock-in. It should plug directly into existing Slack channels, WhatsApp threads, CRM pipelines, and PostgreSQL databases.

### Rule 02: The numbers have to be right
> "When ₦2B moves in a month, one wrong row means someone doesn't get paid."

- **The Principle**: In mission-critical fintech and healthcare systems, "mostly working" is an outage. Stochastic guesses have no place in financial ledgers or medical records. Systems must maintain deterministic consistency, replayable run logs, and atomic transactional boundaries.
- **Application in AI Employees**: AI agents must never possess direct write access to raw database tables or unbounded payment gateways. Every financial action must route through schema-checked, rate-limited tools and require human verification on high-risk tiers.

### Rule 03: Take the boring work off people
> "Automate the copying. Keep a person in the loop for the cases that need one."

- **The Principle**: The purpose of automation is not to eliminate human accountability, but to eliminate administrative toil—chasing missing documents, copying data between spreadsheets, verifying tax IDs, and routing tickets.
- **Application in AI Employees**: We structure workers around the **5-Tier Human Escalation Framework**. The AI worker does 80% of the manual preparation and research, but passes edge cases, low-confidence decisions, and sensitive outcomes to a named human operator with full context.

---

## Related Documentation
- [Overview & Architecture](https://tobilobaodejinmi.com/llms.txt)
- [Services Specification](https://tobilobaodejinmi.com/services/llms.txt)
- [AI Workforce Academy Curriculum](https://tobilobaodejinmi.com/academy/curriculum/llms.txt)