Retool’s New AI-Powered App Builder Lets Non-Developers Build Enterprise Apps

Enterprise-grade software and vibe coding don’t typically go hand-in-hand. Still, increasingly, we’re seeing no-code/low-code platforms that take what they already do best — make developing and safely deploying software inside of an organization easy — and add a generative AI (GenAI) layer on top. Just last week, OutSystems announced updates to its AI-based Mentor service, for example.

This week, Retool is introducing an end-to-end service enterprise application generation (AppGen) platform that lets its users build new tools with natural language, based on their production data, and with their existing enterprise security in place.

While Retool always offered a no-code experience, the end product was always code, so there was always room for professional developers to take over if needed and bring a Retool-based application into their stack. That made the service a bit different from some of its competitors, and as Retool CEO David Hsu told me, only a few years ago, maybe 10% of Retool users were non-developers. Today, that number is closer to 50%, so it makes sense for the company to double down on this shift.

“I was a skeptic of no code/low code for a very long time,” Hsu said. “We didn’t want to call ourselves low code because engineers hated it. But now I can. AppGen is the secret to low code. When you think about AppGen, it basically is low code if you think about it. But I think the technological advancements of LLMs have made it — for the first time — actually possible for non-engineers to go build software.”

But to do this, Retool is taking a slightly different approach from its competitors. Hsu believes that for enterprises to be able to use a tool like this and feel confident in the results, there have to be some guardrails. These include security policies and data governance, of course, but at the core of Retool’s new service is the ability for the professional developers inside the company to create and define the building blocks (Retool calls them “semantic objects”) that these non-developers can then use to build their own applications.

In practice, this means the developers can create a canonical way to display customer data, for example, with the database queries and user interface set by them, and make that available to all users.

On the security and governance side, any Retool app inherits the organization’s single sign-on (SSO), role-based access control (RBAC), audit trails and compliance frameworks.

“I think what’s really interesting now about generative AI or large language models [LLMs] is that they are quite good at some things, but very bad at other things. What they’re very good at, I would say, is the creativity, the speed at which you could prototype,” Hsu explained when I asked him why the company took this approach. Yet while the models keep getting better, they are not deterministic so you can’t really guarantee that the results will always be the same. So for some parts of the stack, including security, the governance layer and data connectivity, LLMs are not the right answer.

“What’s really interesting is that what we have spent the last six, seven years building, is exactly what AI is bad at,” Hsu said. “No developer would probably want an LLM writing their security stack. That feels very dangerous, right?”

The development experience consists of a chat interface on the left and then a preview of the application on the right. Users can easily take manual control, too, with the help of an inspector that provides details and the ability to make edits for every item in the generated application. At the end, after all, Retool generates code.

Once the application is complete, users now also have the option to host it in their own cloud, but managed by Retool.

In a recent survey of about 10,000 companies worldwide, Retool found that almost half of the non-engineers it queried said that they believe they can now build software (and as expected, the majority of organizations — 66% — now have AI productivity mandates).

“AI has dropped the bar to prototype, but the bar to ship hasn’t moved,” said Hsu. “Most AI tools generate code that still requires extensive integration work to become production applications. Organizations are discovering that their biggest AI productivity gains come from platforms that understand their existing data, security and infrastructure from Day 1.”

Indeed, that seems to be the tenor in the industry right now, whether that’s about low-code/no-code tools or developers who use tools like Cursor and Windsurf. As these tools allow more employees to write code, the bottleneck going forward won’t be code generation but testing and integration, which Retool is working to shortcut by choosing the building blocks approach and doing a lot of that work upfront.