A New Company Wants to Keep AI Coding Tools Independent From Big Tech

Greylock Partners, a major investment firm, has backed a startup called Niteshift that was founded by former engineers from Datadog. The company is positioning itself to provide the underlying infrastructure for AI coding agents — software that can write code, test it, and make improvements largely on its own, according to Greylock's announcement. The key idea behind the investment is that this infrastructure should not be controlled by Amazon, Google, Microsoft, or any single AI lab.

The investment was announced on 10 June 2026. Financial terms were not disclosed publicly.

What Niteshift Does

Think of a coding agent as a self-driving developer. It is not just a feature that suggests your next line of code as you type — it is a programme that works independently: reading your codebase, writing entire functions, testing them, and proposing changes through pull requests. It can keep working for hours or days on a single task, and it might use multiple tools along the way.

Because these agents run for long stretches and do complex work, they need different infrastructure than simpler software does. They need places to run safely, a way to keep track of what they are doing, ways to manage passwords and API keys securely, and detailed logs showing what happened — so teams can see what went wrong if something breaks.

The founding principle behind Niteshift is that no single company should control this infrastructure. Right now, Amazon Web Services, Google, and Microsoft have all built their own versions of AI agent platforms tied closely to their services. If you use Microsoft's version, for example, you are locked into using their tools and their preferred AI models. Niteshift's founders believe teams should be free to switch between different AI models, move their work to different cloud providers, or manage their data where they need to — without being stuck with one company's system.

Learning From an Earlier Success

Datadog, the company where Niteshift's founders came from, built its business by convincing tech teams that they needed a single, unified view of everything happening across their systems — rather than cobbling together ten different monitoring tools. Datadog's insight was that fragmented approaches do not work at scale. Companies were willing to pay for one platform that handled it all.

Niteshift is making a similar bet: teams running AI coding agents at scale will want one coherent platform to manage them, not a patchwork of scripts connected to different AI APIs and cloud services.

Datadog itself has already started building agent-monitoring tools for this reason. The company created something called an Agent Console that tracks what its own internal AI assistants are doing — logging their actions and measuring their performance. Even Datadog, a company with years of expertise in monitoring, found that existing tools were not quite right for the job. That gap between what exists and what is needed is exactly what Niteshift is trying to fill.

Why Now

The first wave of AI coding tools — like GitHub Copilot — were easy to add to your existing setup. They worked like any other internet service: your code editor sends a request, gets back a suggestion, done.

The second wave, with agents that work independently over hours, immediately raises new practical questions. Where does the agent run code safely, so it cannot harm the rest of your system? How do you keep track of what the agent is doing if it is paused and then picks up again later? If an agent spawns a long process running in the background, who bills for that compute?

We have been through a similar moment in technology before. Around 2014, when containers — a way of packaging software — became practical for real work, teams discovered that containers solved one problem but created many others. Nobody had the tools to manage where containers ran, how they talked to each other, how to keep secrets safe, or how to figure out what happened when something broke. Over the next few years, Kubernetes and other systems emerged to solve these new problems.

Coding agents are the new container. They solve one problem — how to get AI to write code autonomously — but they create new problems underneath. The infrastructure to manage them safely, at scale, is still being built.

The Independence Question

The timing also reflects real business pressure from the big cloud companies. Amazon, Google, and Microsoft have all released their own versions of agent platforms, and each is designed to keep you using their own services and AI models. If you start with Microsoft, it is easier and cheaper to stick with Microsoft. That is by design.

An independent platform like Niteshift competes by offering freedom: the ability to choose different AI models from different companies, to move your work between cloud providers, or to use cheaper, open-source AI models. The bet is that this flexibility is worth paying for.

However, there is a real challenge to this strategy. If one AI model becomes so good that nobody wants to switch away from it, the whole "we will not lock you in" argument becomes weaker. The counter-argument — and the reason Greylock seems to have invested — is that AI models are improving so quickly, and the capabilities of different models are changing so fast, that any team betting everything on a single model will regret that decision within a year or two.

It is worth noting that Greylock's announcement does not give much detail about how Niteshift actually works — what it is built on, how it isolates code execution, how it handles different AI models. Infrastructure companies typically keep these details quiet until they have signed paying customers. That does not mean the company is hiding anything problematic; it is standard practice. But it does mean we cannot yet verify whether the approach actually works as described.

What Actually Matters

If you are a technology team thinking about this space, the practical questions are: What is the actual interface for building on Niteshift? Does it work smoothly with existing AI coding tools? How does it actually keep code safe when it runs agent tasks that could potentially modify your repository or trigger automatic deployments? And where does the monitoring data go — into your existing tools, or somewhere new?

These details will determine whether Niteshift becomes something teams build on or something one of the large cloud companies eventually buys. The founding team's experience at Datadog — a company that took something as messy as infrastructure monitoring and made it work reliably at scale — is the strongest signal we have right now. Whether they can pull off the same trick with AI agents is the question the next year will start to answer.

The infrastructure underneath AI agents is a genuine problem that needs solving. The teams that have spent years managing complex systems at scale are well-positioned to build it. Niteshift is early, its approach has not yet been tested with many customers, and it faces competition from some of the richest companies in the world. But the problem is real, and on current trends, it is getting more urgent faster than the solutions available to handle it.