Frontier labs and hyperscalers like Anthropic and OpenAI deliver innovation but operate without stability, as seen in policy flip-flops and security incidents. Meanwhile, open-source organizations like Open Claw and Deep Seek offer cost-free models with similar quality, closing gaps in utility and safety. This leaves CTOs, CIOs, and engineering leaders questioning how to maintain reliability and control costs when hyperscaler pricing is unpredictable.

Engineering leaders at Big Tech companies learned that chasing trends like "tokenmaxxing" leads to unsustainable spend and burnout. The pendulum is swinging back toward sober approaches, with companies like Meta reinvesting in engineering culture. Organizations should modernize infrastructure to empower teams to sustainably iterate on AI tools at scale.

The future empowers people and agents to work seamlessly together. Agents need access to real infrastructure and data under the same guardrails as human teams — credentials granted only when needed, actions observable and auditable. Systems must swap in latest models without losing custom work, avoiding single-provider lock-in.

Building resilience starts with accepting models will change. Leaders should embrace flexibility to adopt hyperscaler or open-source models as needed. Eliminating long-term vendor lock-in allows organizations to adapt without rebuilding. Sumeet Vaidya, CEO of Crafting and former engineering leader at Meta, Uber, and Discord, advises adopting this approach today.