Engineering Strategy: Reproducible Decisions via Frameworks
Build engineering strategy through explore-diagnose-refine cycles, using systems models and Wardley Maps for validation, as shown in Uber migrations, Stripe API deprecations, and LLM adoptions.
Strategy Creation Process
Good engineering decisions scale through strategy, which systematizes choices for engineers and executives alike. Start by assessing usefulness: strategy shines for complex, ambiguous problems like migrations or tool adoptions, not routine ops. Anyone can contribute—engineers via analysis, execs via policy—but write concisely (1-5 pages) when aligning large teams or navigating change.
Follow a six-step cycle: (1) Explore constraints and options via stakeholder input and data; (2) Diagnose root causes using causal models; (3) Refine hypotheses iteratively to avoid waterfall pitfalls; (4) Set policy with clear rules like 'deprecate APIs after 12 months'; (5) Run operations to execute and monitor; (6) Make readable with visuals and summaries. Bridge theory (e.g., systems thinking) to practice by modeling real impacts, like velocity gains from LLM tools.
Evaluate strategies by testing assumptions early and measuring outcomes against goals—strong ones predict behaviors and adapt to feedback.
Refinement and Modeling Tools
Refine strategies iteratively: test via simulations (e.g., 'what if we onboard services too fast?'), avoiding rigid plans. Use systems modeling to diagram feedback loops, stocks/flows, and leverage points—e.g., model LLM impact on developer velocity by plotting adoption curves against productivity sinks like context-switching.
Apply Wardley Mapping to visualize component evolution (genesis to commodity) and dependencies: map service orchestration (Uber 2014) or LLM ecosystems (current) to prioritize custom vs. buy decisions. These tools expose blind spots, like over-investing in custom tools when commoditization looms.
Improve via practice: study cases, collaborate with peers, and iterate drafts.
Real-World Applications
Uber (2014) migrated services via onboarding models balancing velocity and stability, using Wardley Maps to evolve orchestration from custom to leased.
Adopt LLMs strategically: model DX gains (e.g., 20-50% velocity boost) against risks like hallucination; prioritize low-hanging onboarding like code review agents.
Private equity transitions: model seniority mix to sustain output amid headcount cuts.
Other cases: Control user data access via tiered policies; decompose monoliths only if modeling shows congestion relief; at Calm (2020), resource product-engineering projects with dedicated pods; Stripe deprecated APIs (~2016) via phased sunsets with models tracking adoption/dropoff; built Sorbet (~2017) for type safety in Ruby; integrated Index acquisition (2018) via tech convergence plans.
AI for Strategy Acceleration
Leverage LLMs as co-writers: collaborate on drafts (prompt with context), review for gaps (e.g., 'check causal links'), generate systems models (input variables/outcomes), and Wardley Maps (describe components/visibility). Foundations: treat AI as junior collaborator—provide structure, iterate outputs. Next: chain tools for full strategies from exploration to visuals.