You Bought the Agents. Now Build the Factory.
Your developers are already using AI coding tools. Claude Code, Cursor, Copilot, probably a few others depending on the team. Leadership sees competitors shipping faster and starts asking why you aren’t. The answer is almost never about the tools. You already have those.
The missing piece is the infrastructure around the tools. Uber calls it a software factory, and after publishing more detail about how it works than any other company this year, the pattern is worth paying attention to.
A coding tool makes one developer faster inside one session. That’s genuinely useful. It doesn’t solve four problems that show up the moment you try to run dozens of agents across an engineering organization.
The first problem is where agents run. A laptop sleeps when its owner does. Tasks that need to start from a broken build, a calendar trigger, or an overnight alert can’t run on someone’s personal machine. Uber describes its shift to managed cloud environments as its core strategic move in AI-assisted development. DoorDash made the same move for a different reason. A developer’s laptop holds SSH keys, VPN credentials, and authenticated access to production systems, and handing all of that to an autonomous agent creates a significant exposure that cloud-managed environments can contain.
The second problem is consistency. Give a hundred developers a coding agent and you get a hundred configurations. Uber found this out the hard way, discovering that engineers had built hundreds of duplicate skills across repositories with wildly varying quality standards and missing security rules. The fix is writing each shared practice once and distributing it across every agent in the company.
Cost visibility is the third problem. Uber spent its entire annual AI budget in four months this year. Individual tool licenses tell you what seats cost. They don’t tell you what any given merged pull request cost to produce. Without that number, the conversation about what to automate and what not to automate has no foundation.
The fourth problem shows up once the first three are solved. More code output means more code review. Spotify saw a 76% jump in pull requests after rolling out AI coding tools across its engineers. Across companies measuring this broadly, review time increased by 441%. Output goes up while the review pipeline stays the same size.
Building a software factory addresses all four. Uber’s approach uses six building blocks, including model gateways, shared skills libraries, and cloud execution environments. None of them is a coding agent. You buy the agent. You build the setup around it.
The right starting point is the work nobody wants to do by hand. Repetitive tasks with clear success criteria and low cost of failure. At Uber, that was automated code migrations. At Spotify, background agents handled about 1,800 downstream pipeline changes for a single dataset update, work that would have taken a team weeks to complete manually.
The goal isn’t to replace your developers. It’s to take the grind work off their plates so their judgment goes where it actually matters.
Want to explore how an AI-native development process could work for your team? Let’s talk.