Starbucks is building its own AI tools instead of buying them, reportedly to trim a ~$400M annual software bill. Three years ago Azati made the same bet – but turned its own 300-person company into the test lab. The rule: we don't sell an AI practice we haven't survived ourselves. Build vs. buy is rarely a clean line; it means running and paying for two systems until the new one can safely replace the old.
The line is moving
For two decades the default was buy – building in-house was slow, costly, risky. AI-assisted development is shifting that line toward build, which is why the "just buy it" reflex is no longer automatic.
Savings aren't a snap
Until the new system fully replaces the incumbent, you run both and pay for both. Build vs. buy is a transition to manage, not a switch to flip – the payoff comes after a successful cutover, not on day one.
The failure lesson
Starbucks had already retired a hyped AI inventory tool in 2026 after it struggled across thousands of stores, sending staff back to manual counts. The real risk isn't spend – it's depending on a replacement that isn't production-ready. Fix the process first, then automate, then migrate deliberately.
That's why we built our platform on ourselves first. How Azati de-risks it: AI Consulting for an honest build-vs-buy assessment, Legacy-to-AI Modernization instead of rip-and-replace, and Managed AI – because AI is an operating model, not a one-time build.
FAQ
Adapted from a LinkedIn post by Andrew Babkin at Azati – read the original. Sources: Bloomberg, Fortune, TheStreet.
