A platform strategy case study: how a Developer Experience team moves an organization from manual infrastructure configuration to an AI-agentic, intent-based deployment model, without trading reliability for velocity, and without asking skeptical senior engineers to trust a black box.
Grounded in Yahoo as a real, public example of the setting: their Platforms org runs Cloud Runtime (Kubernetes), Security, and Messaging as the underlying capabilities a developer has to glue together to ship anything, per their public Sr. PM, Platforms job posting. Everything past that is my own analysis, not Yahoo's internal roadmap or proprietary information. I have no visibility into their actual architecture or metrics.
Engineers write code fast and then lose days hand-gluing four disconnected platform capabilities (Provisioning, Security & Compliance, Observability, Traffic Management) to get a simple microservice into production. This is a 6-month plan to replace that manual "glue" with an AI intent layer, internally named Shepherd: a developer describes what they're deploying, however they already have it (a one-line blurb, a POC repo, a legacy service to modernize), the agent shows back what it understood and composes it from pre-approved, human-authored modules, and a person still approves every change before it ships.
| File | Contents |
|---|---|
| docs/01-opportunity-brief.md | The gap, personas, opportunity, hypothesis, flagged assumptions |
| docs/02-prd.md | Main deliverable: Problem Formulation, Solution Definition, Technical Implications |
| docs/03-event-model.md | The intent to production mechanism, walked command by command |
| diagrams/ | Mermaid sources for the path to production (before/after), the event model, and communication flow, plus the success-metrics chart (light/dark SVG) |
| mockups/index.html | Shepherd's GUI, inside a platform catalog: browse services, self-service actions, all three input modes, confirm, PR review, service scorecard, audit log |
| mockups/tui.html | Shepherd's CLI/TUI: same three input modes as a terminal session, k9s-style persistent context bar |
| appendix/ai-prompts.md | Representative AI prompts used to accelerate research and pressure-test trade-offs |
- Maya: Product Engineer, Tier-2 service owner. Wants to ship without becoming an infra expert.
- Dev: Senior Platform Engineer, the skeptic. Wants to see exactly what automation will do before it does it.
- Sam: Platforms Lead. Wants velocity that doesn't spend the reliability budget.
Frameworks applied directly (opportunity framing, jobs-to-be-done personas, event modeling, low-fidelity wireframes). Benchmark figures (DORA change-failure rates, Copilot acceptance rates, platform engineering adoption stats) are sourced and cited inline in the PRD. Anything without a citation is explicitly flagged as an assumption or illustrative placeholder, not asserted as fact.
- Read the PRD first, it's the main deliverable.
- Open
mockups/index.html(GUI) andmockups/tui.html(terminal) directly in a browser, same underlying flow, two surfaces.
All rights reserved.