Keeping AI-Built Apps Inside Company Rules

People across a company can now build working apps with AI, and those apps still have to follow the same security, compliance, and cost rules as everything else. Cory O'Daniel, CEO and co-founder of Massdriver, says operations, security, and compliance teams can load the tools they already use. A person then draws what they need on a diagram, and Massdriver builds it from those approved tools.

In this episode of the Web Talk Show, I talked with Cory O'Daniel, CEO and co-founder of Massdriver. Massdriver is an internal developer platform, a tool that helps teams set up cloud resources inside the company's rules. A developer may know they need a database, but cost, security, and legal requirements sit with other teams. At a smaller company, figuring that out can take a week. At a large one, it often becomes a ticket that might get done today or in three months.

Cory has been in the cloud since the early 2000s, when AWS was just getting started. He began in healthcare data centers, then moved to California and took his first software job. His boss asked if he used to work in a data center, then announced that Cory would lead the team's move to the cloud. That work later got called DevOps. For about 20 years he built platforms so developers could get what they needed without waiting. He later worked as a consultant for Google and saw companies with 4,000 developers facing the same bottleneck he had seen at firms with 50 to 100 developers.

He left to start Massdriver about six or seven years ago. After 173 investors said no, the company got the last open spot in a Y Combinator batch.

The idea clicked when a senior platform engineer at a large ad company described infrastructure the way he would on a whiteboard: a box for the database, a box for the app, and a line between them, with security and cost assumed to happen in the middle. That's the simplicity Cory wanted. Teams drop a database on a canvas, answer a short set of questions they actually know, and connect it to the app. The system then sets up firewall rules and permissions from a contract the company already defined.

One cost example stuck with me. If a database has to be up 100% of the time, and you later drop that to 99%, you might expect the cost to drop by 1%. Cory said it can drop by about 25%, because getting close to 100% is much harder than getting to 95%.

They designed Massdriver so even the least experienced engineer on a team can act on their own, inside the company's rules. Cory says that engineer is now Claude. If only senior people can change the cloud setup, work piles up on them and the team slides back into waiting on tickets.

He has seen this pattern for a long time. In the 1990s, while he was at Florida Medical Clinic, the help desk got a call that the MRI server was down. He didn't know they had one. It was under a desk, holding patient information, after someone in the office plugged in a machine the vendor had sent. That's citizen development: someone outside IT adding technology to the business. Spreadsheets and no-code tools continued that pattern. Now a language model can stand up an app for someone who has never thought about where it is hosted.

A friend who isn't a software developer told Cory he had an app with about 20 users. When Cory asked where it was running, the friend didn't know. He thought ChatGPT was hosting it. It was on a free Supabase plan the model had set up for him. For a solo project, that can work fine. Inside a company with 4,000 developers, 3,000 databases, and a thousand networks, it becomes a problem if someone asks Claude to build a sales tracking database that stores personal information and the model puts it on a free plan outside the company's rules.

With Massdriver connected, Claude can only choose from the options the company already approved, then ask simple questions a business person can answer, such as how often the tool needs to be up or when the bill would be too high. One customer went from 400 developer seats to 4,000 after the CEO said everyone is a software developer now.

Cory also showed how this can save time. He drew a workshop on a plane, uploaded the drawing to Claude with Massdriver's context engine, and put the session together in about an hour. He said doing that from scratch would have taken about two weeks. Because the building blocks already existed, it used about 9,000 tokens, which he compared to processing the Cheesecake Factory menu.

At the end of the day, more people on a team can now build software. The useful part is making sure that work lands inside the rules security, compliance, and operations already set. If you want to learn more, Cory said to start at massdriver.cloud, and there is a community Slack linked at the top of the site. Their typical customers are larger companies, and the site prices for that. If you have a smaller team and want to try it, he said to ping him on LinkedIn or in Slack.

Watch the full episode on YouTube, or listen on Spotify.