This week, LuminosAI is releasing a new free tier that allows anyone to access the LuminosAI platform to make sure their AI does what they want it to, in just a few clicks.
Here’s how it works:
Describe your agent in plain language. Upload your data, or let us help generate test data. We figure out what to test, build the eval, and run it. You get a clear report showing what passed, what failed, and exactly what to fix, so you can take it straight back to your agent and make it better. Then, we keep watching after launch. It works with every major agent platform, from Copilot Studio to Agentforce.
So, why are we opening up LuminosAI to everyone?
In a few words: More people are building AI, which means more people need help.
Everyone building AI now has to think about risk
For years, we’ve helped governance, privacy, and risk teams answer one question about the AI their companies deploy: “Does my AI actually do what it’s supposed to?”
But the people building AI aren’t confined to data science teams anymore.
Finance teams are building agents that approve payments. HR teams are building agents that screen candidates and answer employees’ questions about pay and benefits. Legal and operations teams are building agents that review contracts, handle customer complaints, and process refunds. These agents aren’t drafting suggestions for someone to review. They’re taking action, often on sensitive data and in front of customers.
The numbers back this up. Gartner projects that by the end of 2026, 80% of low-code platform users will come from outside IT. Forrester found that 89% of development executives are implementing or planning a citizen developer strategy. And KPMG research shows that 52% of organizations plan to rely on citizen developers for most of their custom application development.
Agents accelerate all of it. If anyone can generate an agent just by describing what they want it to do, the number of AI systems inside a company will grow far faster than the number of people qualified to test them.
The builders changed, but the responsibility didn’t
The core question every governance team has to answer is simple: Will this AI do what it’s supposed to do? We’ve been helping governance teams answer that question for years. And it turns out that it’s the same question everyone building AI has to answer the moment they hit publish.
This question matters most when an agent takes actions on its own, talks to customers, produces information teams rely on, or touches sensitive data.
The failures are rarely dramatic, but they can be consequential. An invoice-coding agent assigns the right GL account every time but never sends invoices over the approval limit for sign-off. An IT helpdesk agent verifies identity before resetting a password but never opens a ticket when it gets stuck. A customer chatbot quotes the refund policy perfectly, then applies it to the wrong customers.
Each of these agents would look fine in a quick demo. But they would cause real damage in production, where the output matters most.
Existing tools weren’t made for today’s AI builders
Nearly every tool meant to evaluate AI risk was built for engineers. They give you a place to run tests, but assume you can write code and then leave the hard part to you: knowing what to test for. Asking one AI to grade another and calling it done doesn’t hold up either. A judge model can report a 94% pass rate while checking entirely the wrong things.
Today, the people building agents need two things:
- A no-code way to test their AI before launch, so they know it does what they want.
- A way to keep testing their AI once it’s live, because models and prompts change, agents gain new tools, and users behave in unanticipated ways.
This is what LuminosAI delivers. And I’m excited to make our platform available to anyone who wants to know whether their AI actually works as intended - no coding, no expertise, and no credit card required.
AI evals are becoming table stakes - now is the time to start
I believe evals will become as routine as QA is for software today. No one ships an app without testing it. Soon, no one will ship an agent without an eval. The companies that win with AI won’t just be the ones that let everyone build. They’ll be the ones that make sure everything their people build actually works.
That’s the future we built LuminosAI for.
Get started free at luminos.ai/start-testing