Retool is launching a new platform for AI to automate tasks.
Retool’s Agents will allow users to build agents on top of every large language model on the market
The launch comes as CEOs ask, “How do we get LLMs to actually replace labor?” said Retool’s CEO.
While public debates swirl around AI ethics and safety, Retool CEO David Hsu said the real conversation his clients are having is simpler: “How do we get LLMs to actually replace labor?”
Retool, a platform for building AI applications that works with consulting firms like Boston Consulting Group and companies like AWS and Databricks, has an answer.
On Wednesday, Retool launched Agents, its version of AI agents. Retool’s system will help users build, test, and manage AI agents.
“People are kind of scared to talk about it publicly, and actually, many of our customers even are kind of scared to talk about it publicly,” Hsu told Business Insider of replacing labor.
But as companies debate ways to automate jobs with large language models, they’re also dabbling in the next generation.
On a podcast last year, Salesforce CEO Marc Benioff said, “We’re hitting the upper limits of the LLMs right now,” and that the future lies in autonomous agents. Nvidia CEO Jensen Huang has said he believes we’ll all be working alongside agents and “AI employees” one day.
These virtual assistants don’t just respond to queries or make predictions from patterns; they complete tasks autonomously by breaking down problems, outlining plans, and troubleshooting when they encounter a tough problem. The year 2025 has been billed as the year of agents, as companies from OpenAI to Salesforce to Glean have launched platforms to help people build and manage agents.
Retool’s Agents will not only allow users to build agents on top of every large language model on the market, but also tackle two areas Hsu sees as key to making agents more effective.
One is to build “hyper-specific” agents, Hsu said. These are more accurate than agents set up to do broad tasks like browsing the web. The second is an agent management system called “God View” that allows users to observe what agents are doing at any given point, including recording past behavior. “You can micromanage agents, and they don’t care if they have to be micromanaged,” Hsu said.
Hsu said customers are already using agents for external work like customer service. He gave an example of a company that built an agent to handle customer refunds using three tools: one to look up customers by name, another to find their latest invoice in the payments platform Stripe, and a third to issue the refund.
