The "Odin" platform, which eventually managed all of Uber's stateful workloads, began as a project to containerize sharded MySQL for a single team. This bottom-up approach allowed them to prove the concept and build a working system before seeking wider, more political adoption.
Related Insights
To drive internal change like adopting coding agents, Snowflake's CEO combines top-down goals with bottoms-up enthusiasm. He finds and elevates passionate early adopters—like a founder who fell in love with coding agents—whose influence proves more effective at driving change than executive mandates alone.
MCP was born from the need for a central dev team to scale its impact. By creating a protocol, they empowered individual teams at Anthropic to build and deploy their own MCP servers without being a bottleneck. This decentralized model is so successful the core team doesn't know about 90% of internal servers.
Building a true platform requires designing components to be general-purpose, not use-case specific. For instance, creating one Kanban board for sales, support, and engineering. This thoughtful approach imposes a ~20% development 'tax' upfront but creates massive speed and leverage in the future.
To bypass subjective debates and gain influence, junior engineers can build prototypes for all competing technical approaches. By presenting concrete, comparative evidence after hours, they demonstrate immense value and can quickly establish themselves as technical authorities, accelerating their path to leadership.
When rolling out the Odin platform at Uber, the team intentionally avoided a big-bang launch. They started with their own systems, then expanded to friendly teams, using an incremental approach to build momentum and prove value before approaching more resistant groups.
Dara Khosrowshahi describes a two-step innovation process. First, let teams compete to rapidly "hack" a solution and find product-market fit. Second, once a winner emerges, the organization must systematize and automate that solution through engineering to make it scalable and part of the core platform.
Forcing innovations to "scale" via top-down mandates often fails by robbing local teams of ownership. A better approach is to let good ideas "spread." If a solution is truly valuable, other teams will naturally adopt it. This pull-based model ensures change sticks and evolves.
Stripe's Experimental Projects Team discovered that embedding its members directly within existing product and infrastructure teams leads to higher success rates. These "embedded projects" are more likely to reach escape velocity and be successfully adopted by the business, contrasting with the common model of an isolated R&D or innovation lab.
To overcome widespread resistance and inertia, companies should avoid company-wide digital transformation rollouts. Instead, create a small, empowered "tiger team" of top performers. Give them specialized training and incentives to pilot, perfect, and prove the new model before attempting a broader implementation.
The primary obstacle to adopting a shared platform model is cultural resistance. Teams accustomed to controlling their full stack view shared platforms as a loss of autonomy and a forced dependency. Overcoming this requires building a culture of trust and shared goals, not just proving the technological superiority of the platform.
