Why open source is the cheat code for AI

The frontier of large-language models is shifting daily. GPT‑5.1, Claude 4.4 and Gemini 3 Pro are now routinely outperforming what seemed cutting-edge mere months ago. As commercial AI accelerates, we’re hearing the same question from enterprise leaders again and again: how do we put this power to use?  For those interested in moving quickly, the cheat code is open source.

What we’ve seen firsthand is tens of millions of data engineers, scientists and analysts around the world connect over open source technologies like OpenTelemetry, Prometheus, Linux, Kubernetes and Apache Spark. Across blogs, videos, Git repositories and other public documentation, these advocates engage in unfiltered discussions, share best practices, exchange APIs and dashboards, and more – all on the open web, not in proprietary, walled gardens.

As a result, leading frontier LLMs come pre-wired with knowledge of how to interact with these open ecosystems. They have been trained on thousands of public post-incident reports, detailing how experts respond to incidents with the open source projects. They understand important terminology, like what “scale up the service” means in Kubernetes, to immediately execute on tasks that would likely perplex the many employees who aren’t experts in the technologies.

This helps vendors and customers alike more quickly take advantage of new agentic AI features. We saw this play out directly at Grafana. We were able to build and scale a new AI agent interface in days – all because we didn’t have to train the underlying LLMs to understand and interact with our open source systems.

Of course, nothing in the world of enterprise technology is a panacea. There’s always the risk models could surface bad or inaccurate information from this corpus of documentation. Newer open source technologies also don’t have the same breadth and depth of content. And ultimately, it puts even more pressure on open source vendors to continue to support the communities behind projects critical to the modern IT stack – after all, the community is the force multiplier here, producing the valuable training material for these models.

Under AI pressure, open source is a quick win

AI ambitions are crashing into the realities of adoption. Under pressure from investors, enterprises are trying to move quickly. They want to deliver a chatbot-like experience for engineers, account managers, marketing teams and more to improve productivity and efficiency, and start to generate new growth opportunities.

But nearly half of businesses are concerned they’re falling behind competitors. It’s why more are looking for standalone systems they can quickly adopt and use to support a broad range of use cases. And they want the ability to use their own proprietary, domain-specific data to enhance performance when needed.

This is posing a fresh challenge for the software industry. OpenAI, Anthropic and others have tens of billions of dollars in investment capital, as well as teams stacked with the world’s premier AI talent. Others simply can’t keep up. It’s why even data-focused companies like Snowflake are no longer in the model-building game.

But for some vendors, it’s also not as easy as just connecting their systems to ChatGPT-5, Grok 4 or other leading LLMs the second they hit the market. Often, there’s an integration period, where the technology providers have to spend time, talent and resources training the models on how to navigate their proprietary systems.

That’s not the case with open source – or widely-adopted, open standards like SQL and JSON. The large volume of public content – from how-tos, forum posts and tutorials – on these technologies is now baked into the leading foundational models. Of course, there’s also public documentation on using proprietary software. But it’s often at much smaller volumes, not at the quantity needed to effectively support the reinforcement training most model-builders use.

With the community of open source’s power, it’s more important than ever for vendors to foster dialogue and information-sharing among users. And with more of the daily work interactions taking place in agentic AI interfaces, it’s critical this documentation is robust, accurate and trustworthy. Meanwhile, because of the priority on interoperability, open source helps AI agents deliver a more integrated experience. It’s incumbent on vendors to continue to foster this spirit of collaboration – even as others in the industry reinforce their walled gardens.

In a market that demands results yesterday, open source offers a rare advantage: instant AI readiness. Models already understand these tools and communities already teach them how to use them. That means faster innovation with fewer resources. For CIOs and builders, the real cheat code isn’t just AI. It’s AI, built on open systems.

This article is published as part of the Foundry Expert Contributor Network.
Want to join?