Why We Invested in CrewAI

The conversation around agents has changed. In 2023, most enterprise teams were experimenting with demos, side projects, and developer notebooks. The core question was whether a model could reason through a task and call the right tool. That was an important first step, but it was still far from a production system.

By 2026, the harder questions are the ones that matter. Which agent should have access to which system? How should context persist without exposing the wrong information? How does a team inspect a failed run, replay a workflow, approve a step, or stop a process before it creates downstream issues? What happens when ten agents are operating across sales, support, finance, HR, and operations at the same time?

That is why we invested in CrewAI.

DVC had already been moving toward this thesis

This investment is not a departure for DVC. It is the next step in a thesis that has become more precise as enterprise AI has moved closer to production.

On June 12, 2023, our generative AI investment framework argued that DVC would spend less time trying to predict the winning application layer and more time focused on the AI Ops and tooling layer around foundation models. The idea was straightforward: as foundation models spread, enterprises would need the infrastructure that makes them deployable. Model management, monitoring, privacy, security, and governance were already central to that view.

By April 25, 2024, that thesis had sharpened around GenAI security. The enthusiasm around LLMs was easy to see, but so were the risks: broader attack surfaces, brittle permissioning, role-based access issues, data exposure, and a growing need for observability. Eight months later, on December 16, 2024, our investment in Fiddler placed AI observability on the same path. If enterprises cannot see how AI systems behave, they cannot trust them in production.

On February 3, 2025, our observability work extended that logic into AI agents. Agents do not simply generate outputs. They make decisions, call tools, use context, and take action across systems. That pushes observability beyond model monitoring into behavioral understanding.

By October 27, 2025, our agent-stack view made the missing layer clear: orchestration. Models provide reasoning. Tools connect agents to the outside world. Orchestration manages context, memory, decision flow, handoffs, and multi-agent coordination. CrewAI sits squarely in that layer.

The demo era did not need orchestration. The enterprise era does.

Single-agent demos can hide a great deal of complexity. A prompt, a tool call, and a clean output can look compelling on a conference stage. Enterprises operate very differently. Their systems come with permissions, audit trails, exception paths, approval requirements, data boundaries, and expensive failure modes.

That is where orchestration becomes a product category, not just a technical detail. Enterprises need a way to define roles, assign tasks, coordinate handoffs, manage memory, observe behavior, evaluate outcomes, and govern how work gets done. They also need a framework that business teams can actually understand. If a system only makes sense to a specialized AI infrastructure team, it will have a hard time becoming the way work is automated across the company.

Why CrewAI stood out

There are multiple agent frameworks in the market, and that is healthy. Some are built for highly technical teams that want fine-grained graph control. Some are better suited to assistant-style products. Others are focused more on model tooling, evaluation, or application development.

CrewAI's advantage is its mental model. The idea of a "crew" is intuitive because it maps to how work already happens inside companies. People have roles. Teams have responsibilities. Work moves through stages. Some tasks require tools. Some require memory. Some require approval. CrewAI translates that business logic into a developer-friendly framework without forcing every workflow to feel like low-level infrastructure.

That matters because the next wave of enterprise agent adoption will not be owned by AI engineers alone. Domain experts will matter. Operations leaders will matter. Business users who understand the workflow, the exceptions, and the failure modes will matter. CrewAI's product direction, particularly CrewAI Studio and AMP, reflects that reality.

Commercial traction is also showing up early. In its October 2025 OSS 1.0 announcement, CrewAI reported 1.4B agentic automations, 450M+ monthly automations, and use across 60%+ of the Fortune 500. We do not read that as a vanity metric. We read it as evidence that the product is being pulled into operating environments where governance, reliability, and workflow ownership matter.

The team was another part of our conviction. João Moura has unusual founder-product fit for this market. He built the open-source project before the category had settled on language for itself, and he has kept the framework approachable as the product has moved toward enterprise deployment. We like teams that can earn developer trust and still speak fluently to CIOs, operators, and governance teams. CrewAI has that mix.

João Moura, Founder & CEO, CrewAI

"Dallas VC is exactly the kind of investor founders hope to have in their corner-responsive, insightful, and always ready to help. We're grateful to have Dallas VC as a partner on our journey."

Figure 3: Developer adoption gives CrewAI a credible path into the enterprise.

Star-history chart from Star History, with current repository data from the GitHub API. Open-source adoption is a sourced fact. The DVC interpretation is that this adoption creates feedback loops, community trust, and a path to enterprise standardization.

The product is moving from framework to control plane

CrewAI's early adoption started with the open-source framework. Since then, the company has expanded toward a broader platform for building, deploying, monitoring, and governing agents.

CrewAI's 2026 survey found that 74% of respondents consider agents a critical or strategic priority, while 65% already use agents in production or team workflows. Its Discovery launch introduced no-code agent creation, tool repositories, templates, and a deployment path through CrewAI's Agent Management Platform. Its cognitive memory work adds context engineering, entity memory, user memory, short-term memory, and long-term memory so agents can carry state across tasks instead of treating each workflow as a blank slate.

CrewAI AMP, along with the company's broader agent operations work, is aimed at the enterprise control plane: build with code or no-code tools, deploy to cloud or self-hosted environments, run with RBAC and audit logs, connect tools and agents, observe execution, and manage workflows over time.

Why this fits DVC

DVC invests in enterprise software and AI infrastructure, with a focus on companies that help large organizations adopt new technology safely and at scale. CrewAI fits that lens because it is not just an AI application. It is infrastructure for how agentic work gets designed, coordinated, observed, and governed.

The timing matters too. In our internal conversations, the recurring question was not whether agents would be useful. The question was what enterprises would need once agents began touching production workflows. The answer kept returning to the same requirements: orchestration, memory, control plane, observability, governance, and access for the business teams that actually own the workflow.

CrewAI is building at exactly that layer.

Ravish Ailinani, Partner, Dallas Venture Capital

"CrewAI stood out to us because João and team are building from a rare combination of developer trust and enterprise discipline. They understand that agents become core infrastructure only when companies can observe them, govern them, and put them in the hands of the people who know the work best."

Closing thought

We invested in CrewAI because we believe the next generation of enterprise automation will be multi-agent, workflow-aware, observable, and governed from the start. The companies that define this category will not just help developers build agents. They will help enterprises run them.

CrewAI has the open-source credibility, product velocity, commercial momentum, and enterprise orientation to become that operating layer. If enterprises are going to adopt agents across real workflows, they will need a framework that makes autonomy practical without making it hard to manage. CrewAI is one of the clearest examples we have seen of that future.

Sources

1. DVC, "Generating Value: Exploring Investment Opportunities in the Generative AI Landscape".

1. DVC, "Investing to Secure GenAI".

1. DVC, "Why DVC Invested in Fiddler".

1. DVC, "Observability: Current Landscape and Emerging Trends".

1. DVC, "The State of Agents Today".

1. DVC, "The New People Ops Infrastructure Layer: Why We Believe in Aragorn", used as a style reference for DVC investment writing.

1. CrewAI, "The State of Agentic AI in 2026".

1. CrewAI, "CrewAI Discovery".

1. CrewAI, "How We Built Cognitive Memory for Agentic Systems".

1. CrewAI, "CrewAI AMP, The Agent Management Platform".

1. CrewAI docs for AMP, memory, RBAC, and OpenTelemetry export.

1. GitHub repository data, crewAIInc/crewAI, observed May 17, 2026 through the GitHub API.

1. Gartner, task-specific AI agents in enterprise applications prediction.

1. Deloitte, "Crossing the agentic AI divide".

1. CrewAI, "CrewAI OSS 1.0 - We are going GA".

1. DVC team profile, Ravish Ailinani.

1. The Org, CrewAI leadership profiles for João Moura and Rob Bailey.

‍