Dust 2025 Wrapped: 80,000 Agents, 12 Million Conversations (and other wild stats)
Gabriel Hubert
One product manager edited 560 AI agents in 2025. He's not a software engineer, and he's not alone.
Across Dust this year, something transformational happened. 2025 was the year AI work became team work. More collaboration, more shared agents, more collective building. A one-stop shop of sorts. We're talking about tens of thousands of custom agents, 12 million conversations (that’s one conversation for every person in Bolivia), and a key shift in how teams work.
This is Dust 2025 Wrapped: the real story of how AI got used this year, according to the data.
The Builder Boom
Let's start with the most surprising number: one company configured 7,683 AI agents in 2025. Each purpose-built for specific jobs. They're specialized tools with clear mandates.
Other companies are using Dust agents for:
- A fraud detection orchestrator agent for retail that manages a team of 6 specialized agents to detect if a refund request is a fraud by doing: address check, return distance vs origin, customer history analysis, logistics incidents tracking, payment incidents tracking, text fraud pattern matching.
- A risk assessment agent processing 2.09 million messages, which helped a fintech company run production-grade compliance workflows
- A business analysis agent leveraging our Snowflake connector to handle 65,000+ queries and extract meaningful information from huge amounts of data
In 2025, we learned that AI adoption doesn't look like everyone using one mega-tool. It looks like every team building their own.
And building agents isn’t about code anymore. In fact, most of the people building AI agents on our platform aren’t engineers.
We noticed a product manager built 242 agents. An operations lead edited hundreds more. Product managers, operations leads, and people who know their processes intimately, turned that knowledge into agents that get real work done.
It’s been exciting seeing AI go from "give me an answer in a chat window" to "create something I can share with my team."
The Usage Proof
Okay, so people made a bunch of agents. But are they being used for complex tasks, like running compliance checks and analyzing data, not just casual questions?
The data answers loud and clear: yes.
One company's users sent 1.9 million messages to their custom agents in 2025. Another logged 1.76 million. A third hit 1.18 million.
We're seeing people entirely rebuild how they work around AI. Instead of treating these agents like novelties, our users were treating them like colleagues.
Beyond Chat: The Frame Revolution
But here's where things get really interesting.
In 2025, Dust users learned to create interactive visuals called Frames. Think: dashboards, visualizations, analysis tools, presentations. Things you can edit, present… and most importantly, share.
One company's users created 5,195 Frames this year.
And individual creators had their Frames visited over 100 times. It’s clear these aren’t one-off exports, but these are visualizations being circulated inside organizations.
The Thesis: Democratization is accelerating faster than anticipated
In 2025, people started using AI as a platform, rather than an isolated tool for one-off tasks.
For decades, "building AI" meant hiring a team of ML engineers, collecting training data, and deploying models. Now it means a product manager spending an afternoon configuring an agent that knows your product docs inside-out.
The barrier to building collapsed. Every team is becoming an AI builder because the tools finally let them express what they need without writing a line of Python.
The word democratization gets thrown around a lot. At Dust, what we believe this looks like is the people closest to the problem finally being the ones who solve it.
Looking Ahead
So what does 2026 hold?
This momentum isn't slowing down. In 2026, expect to see:
- More collaboration: Agents will sit in the middle of teams, not simply at the end of a prompt. Human to agent to human. A renewed vision for human-AI collaboration.
- More Frames: AI outputs will be things you can actually use and share, not just text in a chat window. (e.g. dashboards, reports, tools)
- More builders: The people who were power users in 2025 will become platform architects in 2026
The companies winning with AI won't be the ones with the best ChatGPT subscription, they'll be the ones who turned their entire team into AI builders.