iFood: Scaling AI Adoption Across 3,500 Engineers
First member of the DATA & AI department in Marketplace — the largest business unit at Latin America's leading food delivery platform. Capturing, connecting, and packaging AI patterns that emerge organically across 30+ squads.
Latin America's largest food delivery
- 3,500+
- engineers in Marketplace
- 30+
- squads experimenting with AI
The challenge
iFood’s Marketplace — the core food delivery product — runs on 3,500+ engineers. Every squad is experimenting with AI independently: some with Cursor rules, some with MCP servers, some with custom harnesses. The result is a patchwork of isolated wins. Nobody knows what works across the org. Nobody can replicate what one squad figured out.
The AI adoption mandate from the CTO is clear: everyone uses AI, it’s a directive, not an option. But directives without coordination become chaos.
The mandate: capture, connect, package
I’m the first member of the new DATA & AI department in Marketplace. My job isn’t to build AI products — it’s to make 3,500 engineers ship faster by turning scattered AI wins into common patterns.
The adoption flywheel
Input
Squad-level AI experiment
- 01Capture
Enter each BU's active fronts, document what's working and what's blocking. Ads with internal AI coding platform + design tools, Pricing with pricing intelligence tools, Mobile with editor rules + custom MCP servers.
- 02Connect
Link squads doing similar things in isolation. The Ads team doesn't know Pricing solved the same problem. My job is to make them talk.
- 03Package
What proves good becomes common standard: harness configs, skill catalogs, ADRs, MCP servers. Practice becomes adoptable pattern.
Output
3,500 engineers ship AI-assisted code with shared patterns
What shipped
First 30 days — groundwork
- Required:Ads Portal pilot validatedFirst territory of execution. AI-assisted PRD creation, design integration, and review workflow established.
- Required:Working group with DevXMulti-business-unit working group with the architecture enabler team. Presented to leadership.
- Required:Network mappedConnected with 15+ key people across multiple business units and platform teams.
- Required:Baseline instrumentedInstrumented baseline metrics against which adoption will be measured.
The 3-layer architecture I navigate
iFood runs on 3 structural layers plus cross-cutting concerns. My department cuts across all three:
| Layer | What it does | My role |
|---|---|---|
| C1 — Value Generation (BUs) | Where AI becomes money/product | Measure results here — capture+connection appears in BU metrics |
| C2 — Capabilities | Tech stack, platforms, AI tools | Use, adapt, connect existing capabilities; package what works |
| C3 — Enablers | Culture, way of work, disciplines | Human connective tissue — network of AI champions |
My fuel is C2. My terrain is C3. My result is measured in C1.
The iFood playbook: 4 strategies + 1 mandate
iFood’s tech culture runs on accumulated strategies. The AI adoption mandate is the newest, and the one where I operate:
- Engineering resilience culture (est. 2020) — cybersecurity and resilience DNA. Not my responsibility, but my patterns must respect it.
- Platform architecture discipline (est. 2020) — technical and organizational platform thinking. Not my responsibility, but it’s the structure I build on.
- Small-team experimentation model (est. ~2021) — small teams testing theses fast. Phase 2 for me — after I prove capture-connect-package works.
- AI adoption mandate (est. 2026) — my primary vector. The universal directive that everyone uses AI. My job is to make it coordinated practice, not chaos.
Stack
- Cursor
- Claude Code
- AI coding platforms
- Multi-model gateway
- Internal harness frameworks
- MCP servers
- AI automation pipelines