Introducing Discovery

The agent bottleneck is not building, but knowing what and how to build it.

Going from an idea to a working prototype has never been faster. In the world of agent automations, this has resulted in lots of POCs and limited production ROI. The new unlock is knowing what to build and how to build for production. That's the gap that matters most right now, and almost nobody's working on it.

‍

Today we're launching Discovery, a new engine inside CrewAI that helps companies close it.

The right target

‍

The leaders I talk to rarely start with "how do I build an agent". They have a goal, getting real value from AI Agents into production in a way that actually changes how the organization operates. The question is always where to aim.

‍

I've seen what happens when they aim wrong. Most agents that stall don't fail because the technology broke. They fail because the use case didn't match what the business actually needed. The agent works fine. Nobody adopts it. I've watched this happen at company after company, and it's the most frustrating failure mode in this space because it is avoidable.

‍

Getting it right means finding the intersection of three things: what matters most to the business, what you're equipped to do given your tools and data, and what you know will work, backed by real data, not intuition.

‍

Discovery helps companies zoom into that intersection in a way that wasn't possible before.

‍

What exists today

‍

Companies run this process internally or bring in external partners. Either way, it usually means weeks of stakeholder interviews, teams flying across regions, decks stacking up. The output is often useful, a prioritized list of use cases with ROI estimates. But it's a snapshot. It works at a high level but doesn't permeate into individual departments and specific business needs.

‍

It's a photograph taken from very far away.

We took a different approach.

After billions of agentic executions across our platform, we've built a deep understanding of which agent patterns actually work in production, across industries, process types, and at different levels of complexity. We used that understanding to build Discovery.

‍

How it works

‍

Under the hood, Discovery runs multi-signal matching, cohort analysis, and structural pattern recognition against our production data. All you do is tell us about your business. A few minutes later, you're looking at a set of use cases, specific to your company, not generic templates, but real-world processes reimagined based on what we know actually works.

‍

Each one shows what the automation does, the expected impact, the complexity, and how it would work in your organization. These are suggestions matched against patterns that are already running successfully.

‍

You'll also be able to connect your own data sources, internal docs, knowledge bases, existing systems, so Discovery can work with your internal context alongside the external signals.

‍

When I demoed this to business leaders at a $60B global enterprise, at first their reaction actually caught me off guard: "Who told you about this?”

“This is exactly what we need. It's actually making us rethink some of the other initiatives we're already running, the way you're describing these use cases changes how we think about the whole roadmap."

‍

We have customers running Discovery multiple times a day, using it across different business units. It's becoming part of how they plan and iterate on their AI roadmap over time.

‍

The results from early users are clear. More automations getting built. More adoption inside these companies. When people can see what to build, understand how to build it, and trust that the recommendation is grounded in evidence, they build more, faster.

‍

What to build and how to build it is the question that matters now. That's what Discovery solves, not by replacing the work of figuring out your AI strategy, but by grounding it in evidence and making it accessible to every team in your organization, not just the one with the budget and the timeline to run it manually.

‍

Discovery is live today for every CrewAI user.

‍