Niteshift, a coding infrastructure startup founded by two early Datadog engineers, just raised $7 million to solve a problem that's quietly reshaping how enterprises think about AI: they don't want to hand their source code directly to the same frontier labs trying to eat their software markets. The seed round, announced on June 10, 2026, was led by Greylock partner Jerry Chen and included checks from Reid Hoffman, Datadog co-founders Olivier Pomel and Alexis Lê-Quôc, and other prominent investors.

The timing reveals a growing tension in the AI coding market. As OpenAI, Anthropic, and other frontier labs push deeper into vertical software for legal, healthcare, and finance, enterprises face a dilemma: use the best coding agents available, or protect themselves from vendor lock-in by keeping their proprietary code away from companies that might become their competitors. Niteshift is betting there's a third option: a neutral infrastructure layer that lets companies swap between models without rewriting their developer tooling.

What Problem Is Niteshift Actually Solving?

The company's pitch draws directly from CEO Sajid Mehmood's experience at Datadog. Just as Datadog built a multicloud business by selling to e-commerce customers who refused to run on Amazon Web Services while Amazon was simultaneously competing in their retail business, Mehmood argues the same dynamic is now rolling through software. Enterprises want access to the best AI coding agents, but they're wary of handing their source code to companies that might later compete against them in their own industry verticals.

Niteshift's product sits underneath popular coding agents like Claude Code and GPT-based tools. Rather than replacing these agents, the platform routes between Claude, GPT, and open source models on a per-project basis, presenting itself as model-agnostic infrastructure. The company charges per-minute usage rates, like a cloud provider, rather than reselling tokens. This pricing model is a deliberate positioning move: Niteshift is selling infrastructure to enterprises, not selling the agents themselves.

"The thesis is about unbundling," said Jerry Chen, the Greylock partner who led the round.

Jerry Chen, Partner at Greylock

Chen pointed to a pattern playing out across the AI stack as frontier labs expand from model providers into full application businesses. The question is whether enterprises will actually value the optionality of switching models more than the integration depth a single-vendor stack can offer.

How Does Niteshift Compete Against Much Larger Rivals?

The $7 million seed round is modest compared to the capital reshaping this category. Cognition, which builds the Devin coding agent, recently raised $1 billion at a $26 billion valuation. OpenRouter, an AI gateway platform, pulled in $113 million at a $1.3 billion valuation. Amazon Bedrock already ships model routing as part of AWS. Cursor, a developer-focused IDE, has spent years building its user base and may soon be acquired by SpaceX.

Niteshift is entering as a seed-stage challenger to companies already valued more than 100 times larger. The founders' counter is operational. Mehmood and co-founder Conor Branagan didn't just study the engineering organization problem; they scaled through it at Datadog, building infrastructure that handled production telemetry for thousands of enterprise customers. Their pitch to buyers is that AI-generated code will need to be run, tested, and verified autonomously in real production environments, and that requires infrastructure built by people who have done it at scale.

Steps to Evaluate Model-Agnostic Infrastructure for Your Organization

• Assess vendor lock-in risk: Evaluate whether your industry faces direct competition from AI labs offering vertical solutions, and whether code portability matters more than deep integration with a single vendor's stack.
• Calculate switching costs: Determine the cost of rewriting developer tooling if you need to migrate from one coding agent to another, and compare that to the cost of a neutral routing layer.
• Review infrastructure maturity: Examine whether the infrastructure provider has production experience at enterprise scale, particularly in handling autonomous code execution, testing, and verification in real environments.

The skeptic's case is straightforward: frontier labs are not standing still. Anthropic and OpenAI are bundling more orchestration, observability, and deployment tooling directly into their agent products, narrowing the gap that companies like Niteshift want to exploit. A model-agnostic layer only matters if customers actually value optionality more than the integration depth a single-vendor stack can offer.

With $7 million in the bank against competitors carrying nine and ten figures, Niteshift has to prove that wedge holds before its runway runs out. But the signal from who wrote the angel checks is telling. Pomel and Lê-Quôc built the playbook for selling neutral infrastructure into engineering organizations that wanted to hedge against their cloud provider. If the same pattern repeats in AI coding, the winners won't be the agents themselves. They'll be the layer that lets enterprises swap agents without rewriting their developer tooling.

This story reflects a broader unbundling thesis that keeps showing up in seed pitches and enterprise procurement decisions: the model is increasingly becoming a commodity input, and the competitive moat is moving to orchestration, routing, and verification. For the AI coding market, the question is whether that shift happens fast enough for Niteshift to establish itself before the frontier labs close the gap.