One of my favorite business-technology-society-history books is The Machine that Changed the World, by a bunch of guys from MIT — Womack, Jones, et al. In it, they describe the technological and organizational breakthroughs behind what we now know as lean, that revolutionary, begin-with-the-end-in-mind approach that drives agile thinking and end-user centric design, engineering and production processes. It originated radical insights like driving design with production-economies in mind.
If you’re wondering which one of these guys was the scrum master, you probably think the “machine that changed the world” is the one you’re reading this post on. In fact, it’s the one you drove to the grocery store in, the automobile. The book was written in 1990 (yes, I know mainframes already had virtualization by then). That’s where Kanban comes from, the auto industry. Devops and Detroit?
So it’s with no small amount of interest that I read an interview with Randy Mott, following his escape from the challenges of HP to be the CIO of GM. I once met a previous CIO of GM, Ralph Szygenda, who explained that when GM introduced OnStar in 1996, he fought to have the service include two-way cellular communications, to no avail. Another missed technology opportunity.
Mott’s own history as CIO of HP broke some significant ground — along with a lot of eggs, egos, and glass — in driving to consolidate 85 global datacenters into six. Here’s what he said at the time:
“The data centers will provide our business with more dependable, simplified operations. This effort will enable faster delivery of new technologies, services and information and provide room for growth and improved business continuity, while significantly reducing costs.”
Sure seemed like a good idea at the time, and who would have thought that Mark Hurd’s expense reports would have put a stop to the plan?
Now, while consolidation seemed like a good idea in 2006, fewer bigger operations turned out to be necessary without being sufficient. So here’s Mott’s new idea: kill outsourcing. His plan at GM is to bring 90% of IT spend and headcount inside the company, a complete inversion of their current 90% outsourcing approach that feeds HP/EDS, IBM, CapGemini, Wipro et al $3B. He did something like this at HP, cutting IT headcount by half to reach 10,000 employees, 90% on staff.
No doubt, consolidation drove a lot of virtualization. One might also make the argument that consolidation + outsourcing = cloud. But this is where the disciplines of lean engineering re-enter the picture. By in-sourcing his technical talent, Mott is making a fascinating bet on agile IT engineering right where his bosses missed the boat (or the car) on agile automotive engineering. We all know a thing or two about how outsourced IT works:
“When the business says ‘go,’ then that means we start working on a contract, we don’t start working on a project,” Mott says of the current outsourced model.
I would argue the connection between outsourcing and cloud ran through the business unit, rather than IT consolidation. Lines of business were first frustrated by gridlock in IT organizations that came from the unintended consequences of infrastructure sprawl. Then, they were frustrated by IT being tied up in managing outsourcers and not working on anything new that the business needed. Next, they became frustrated by the tone-deafness of outsourcers who knew how to use the same vocabulary as IT, but had little vested interest or insight into really driving competitive advantage for their customers. Next thing you know, the business is hiring its own developers, and slapping down the AMEX to rent some AWS EC2.
Business being what it is, change is the only constant, and the cloud is uniquely suited to absorb and change. But it’s not virtualization, or rentable services that makes change work — it’s agile engineering. And as Mott’s planned transition at GM recognizes, that agility works best when you drive it end to end, just as kanban and just-in-time engineering ensured that customer requirements changed everything all the way up through the tooling of the doors and the design of the bolts. Agile isn’t just user-facing use cases with goals; done right, it drives all the way down to the metal.
Now, when we combine the tooling and the development process, we call it DevOps. OpenStack Cloud infrastructure enables this with open source, delivering transparency from application down to metal. Open Source also helps in providing ways to standardize and adapt tooling top to bottom using independence from external vendors that lets you capture competitive advantage and control your destiny. As OpenStack Quantum and software-defined networking make their way into the wild in the Folsom release this September, you’ll see deeper and deeper alignment between infrastructure tooling and development, such that DevOps will become the norm.
I’d be willing to bet that Mott’s new $3B IT operation at GM will succeed or fail on its ability to integrate agile DevOps strategies with an agile infrastructure. Ideally, GM will remember the lessons of the last time it got an ass-whupping for ignoring the lessons of agile engineering.