Anaconda Buys Metaflow Maker to Tackle AI Code Quality Crisis

Anaconda, a provider of AI-native development platforms, announced Wednesday that it is acquiring Outerbounds, the company behind Metaflow—an open source orchestration framework for AI and machine learning that originated at Netflix. The acquisition aims to equip enterprise teams with a controlled pathway from initial AI experimentation through to production deployment, positioning Anaconda as a comprehensive platform for AI-native development.

The move underscores a critical inflection point in enterprise software development. According to Anaconda's analysis, AI-generated code now represents nearly half of all new code written in enterprise systems. However, that code introduces 1.7 times more defects than code written by humans, and 80% of dependencies suggested by AI coding assistants carry known security vulnerabilities.

Agents are introducing 1.7x more bugs into the software.

David DeSanto, Anaconda CEO

Anaconda CEO David DeSanto explained the tension organizations face: "There are a lot of organizations saying to us: we have to trade off between the velocity of an agent and the quality of a human." He elaborated that "Agents are introducing 1.7x more bugs into the software, which means humans are trying to fix that, which means they're almost not getting the value of leveraging the agent to start with."

The bottleneck is somewhere else, now

DeSanto argues that the primary constraint in the AI era has shifted. Rather than code generation itself, the real bottleneck lies in managing and governing all dependencies that code relies upon—doing so reliably across distributed systems, with reproducibility and security as foundational requirements.

In a press release, DeSanto stated: "The future belongs to AI-native development, where the AI model is the core of how applications are built, not something bolted on at the end. The problem enterprises face today is that delivering on that vision requires stitching together tools, platforms, and governance components that were never designed to work as one, nor to even work with AI. Until now, no other platform has spanned the entire AI-native development lifecycle."

Speaking with The New Stack, DeSanto added: "We don't want to just be for data scientists. Data scientists are becoming AI engineers now. Software developers are being asked to write code that works with AI models. Realistically, everyone's becoming that AI developer."

Netflix DNA, enterprise scale

Metaflow has established its credibility in rigorous production environments. Originally developed inside Netflix to manage production-scale AI and ML workloads, the framework is now deployed by organizations such as Realtor.com, GE HealthCare, and Warner Bros.

Outerbounds has evolved Metaflow into a comprehensive enterprise offering. According to Ville Tuulos, CEO and co-founder who previously oversaw AI and ML operations at Netflix, the platform manages orchestration, compute scaling, and deployment across any infrastructure without binding customers to a single cloud vendor.

There's a big responsibility aspect that oftentimes gets forgotten in this AI mania.

Ville Tuulos, Outerbounds CEO

Tuulos explained: "Netflix has this cultural value of freedom and responsibility. It's very useful to have enough freedom to choose the best tool for the job. But there's a big responsibility aspect that oftentimes gets forgotten in this AI mania."

Anaconda's cloud-agnostic approach aligns with its broader market positioning. The company serves more than 50 million users and has accumulated 21 billion downloads by functioning as the foundational platform for Python-based data science and AI development, supplying secure packages, validated dependencies, and reproducible builds.

Secure by default, not by bolt-on

Both companies emphasize that security is embedded into the platform architecture rather than added afterward. Tuulos told The New Stack: "Our platform always deploys in the customer's own environment, which is actually somewhat different from any other SaaS services. For all of our customers, everything runs securely in their own cloud or on-prem. Doing that is much harder — but there's a lot of value in it."

The acquisition bridges a longstanding gap by extending Anaconda's trusted foundation through to production orchestration—a capability that has historically required teams to integrate multiple disconnected systems.

For Tuulos, the strategic fit rests on shared engineering principles. In a statement, he said: "What makes this combination so powerful is a shared commitment to Python, reproducibility, and software engineering best practices. Together, we can give data scientists and AI engineers everything they need to move from secure environments to production-grade orchestration, and turn AI innovation into real, measurable outcomes."

When the model is the core

The acquisition also reflects fundamental structural differences between AI-native development and conventional software engineering. In AI-native applications, the model functions as the central component—not as a feature layer added to traditional code. The surrounding infrastructure exists to support the model: supplying data inputs, handling outputs, managing dependencies, and ensuring everything maintains security and reproducibility.

AI agents are providing the code, but this idea of providing the outer bounds.

Ville Tuulos, Outerbounds CEO

Tuulos explained to The New Stack: "The entity producing the code in the middle is not a data scientist anymore — increasingly it's AI agents. So, the AI agents are providing the code, but this idea of providing the outer bounds, those boundaries in an enterprise environment within which you can run this code with confidence — that is more relevant than ever." This rationale also explains Outerbounds' company name.

Human developers continue to participate in the process, establishing requirements and making architectural choices. However, the sheer volume and complexity of code flowing through enterprise systems has now exceeded what manual review can handle.

One stack, your infrastructure

Anaconda states that the combined platform delivers workflow orchestration, compute management, experiment tracking, and enterprise governance through a single integrated stack, operating on infrastructure that organizations already own and manage.

Anaconda has committed to maintaining Metaflow as an open source project, with its engineering team contributing to the framework alongside the commercial offering. This approach reflects Anaconda's longstanding commitment to open source stewardship across the data science and AI landscape.

Source: The New Stack