Introducing Niteshift: the full-stack cloud for coding agents

Coding agents are redefining what is possible. Every month brings another incredible demonstration of agentic engineering: a new browser rendering engine built from scratch, a C compiler that builds the Linux kernel, a full port of Bun from Zig to Rust, done in a week. The raw capability of what agents can write today is no longer in question.

But shipping velocity is not scaling at the same rate. Coding agents are taking on tougher, longer-horizon tasks, yet most teams are seeing only a fraction of the promised productivity gains. To work autonomously, coding agents need a way to verify their own work. At a minimum, that means giving agents a real environment with all the context and tools an engineer would use to test their code. Without that, agents cannot close the loop themselves, and the productivity gains that should compound across the entire engineering org get absorbed back into manual review cycles instead.

Why agents can't close the loop

Real applications are more than a repo. They depend on Docker services, databases, credentials, feature flags, background jobs, and seeded test data. Without a full runtime, agents can write code but they cannot prove the checkout works, the chart renders, the bug is fixed, or the regression is gone. They need the full verification loop: integration tests, browser automation, logs, and CI.

And once agents can verify their own work, a second constraint appears. Agents that can close the loop run for hours before needing human review. Engineers need to run dozens in parallel to stay productive, far more than any laptop can support.

What coding agents actually need

Engineers today are rationing their git worktrees, struggling to run more than a few dev environments in parallel, even with a $5000 128 GB RAM MacBook. Production software solved this by moving to horizontal scaling in the cloud. Coding agents need the same structural transition.

Agents need the entire application stack running, with enough elastic capacity to run many of them simultaneously, accessible to everyone on the team regardless of their technical setup, and free from any bet on a single coding agent vendor at a moment when the best agent changes every few months.

Our obsession with shipping

Conor and I have always loved shipping. We met at Datadog, where Conor was the number one committer in Datadog's original monorepo, and I led teams that launched a dozen new products. The first time we played with Claude Code, we both had the same reaction: this is going to let us ship so much faster. But the smarter coding agents became, the more we kept hitting the same verification bottleneck. The gap is now infrastructure, not intelligence.

Closing that gap requires a platform that can adapt to the incredible diversity of real production software. Every stack is different. Every codebase has its own auth flows, service dependencies, and undocumented quirks. At Datadog we spent a decade building the observability and verification layer that had to work across all of it. The frontier labs are pushing model intelligence forward faster than anyone expected. We are building the platform that translates that intelligence into reliable, production-ready software.

Introducing Niteshift

Today we are launching Niteshift: the full-stack cloud for coding agents.

Give Niteshift a prompt, bug report, Linear ticket, prototype, or half-finished PR, and it returns a pull request with evidence that it works.

Full-stack runtime. Real applications need more than a cloned repo to run. Niteshift's setup agent automatically configures complex dev environments to run end to end in the cloud, including authentication and seeding test data. Every task runs against the full application stack, containerized services, persistent storage, auth, queues, workers, all wired together the way the application actually runs.

Standard Bots runs its full motion planning and robot control stack inside every Niteshift task. "Niteshift is the only platform we found that can spin up the services needed to simulate our robots end to end. It accelerates our engineers by letting them run parallel agents in production-like environments, while designers can merge PRs with zero setup." — Evan Beard, Founder and CEO, Standard Bots

Unlimited concurrency. Once agents have a real environment to work in, the only remaining constraint is how many can run simultaneously. Niteshift spins up as many parallel environments as needed, 24/7, without a laptop in sight. Agents run for hours without getting blocked.

Listen Labs points Niteshift at eval scripts and lets agents iterate until the metric moves for autoresearch-like optimizations. "Our engineers point the agent at an eval script and let it run for hours optimizing our whole codebase on Niteshift." — Florian Juengermann, Co-Founder Listen Labs

Collaboration by default. Niteshift agents are accessible from everywhere your team works: Slack, Linear, GitHub, web, and mobile. Agents can be triggered by any engineer, PM or designer in the organization, automated on a schedule or in response to external webhooks. When the environment lives in the cloud, access is no longer limited to engineers with a working local setup.

Any frontier agent. The best coding agent changes every few months, and is different for different tasks. Niteshift is built around that reality. Teams define the environment, tools, policies, and path to production. Engineers bring any agent: Claude Code, Codex, OpenCode, Pi, or whatever comes next. The environment, tools, and automations stay the same regardless of which agent is running.

Elicit built The Line, an internal orchestration layer that fires off parallel Niteshift tasks via API. "We evaluated several background agent platforms and picked Niteshift. It produced the best results on our evaluations, both because of the agents available and the tooling around them." – James Brady, Head of Engineering, Elicit

The new measure of AI-native software teams

Many companies track AI adoption with token-usage leaderboards. This is backwards. Instead, the best measure of a team's adoption of agentic engineering is the portion of shipped software that is created autonomously with agents. This is the best encapsulation of the quality of your agent tooling & infrastructure, and as this metric increases, your backlog shrinks, bugs are fixed sooner, and the roadmap gets more ambitious and ships faster.

The most AI-forward teams have been pushing this number hard. Stripe's Minions ship more than 1000 PRs per week. Shopify's River merges 1 in 8 PRs across the entire company. Ramp's Inspect has grown from 30% to 70% of merged PRs in just the last few months. All three built their own in-house infrastructure to get there, with dedicated engineering teams to keep up with the fast-moving frontier. Even with AI assistance, it takes months of work to get right – if you have strong pre-existing developer infra to build on top of. Most teams do not have this.

Which is why we built Niteshift. The top teams on our platform set up their environments in under an hour, and within a few weeks ramped up to merging over 75% of their PRs via Niteshift. The most prolific engineers consistently merge 100+ PRs per week, singlehandedly. And since work in Niteshift is legible to the whole team by default, the best practices naturally diffuse throughout the company.

Raising our ambitions

While the results for our early customers have already been transformative, there's still so much more to build. Today, the best agent dev environments look broadly similar to human ones. But agents are already superhuman at tool use. Codex composes bash one-liners that chain grep, awk, sed, and find to operate across hundreds of files in a single pass, exhaustively and precisely in a way no engineer would ever write by hand. That's a strong signal that agents are capable of far more, if given tools purpose-built for their unique abilities. This is what we're building at Niteshift: the cloud for agents that raises the ceiling on what we can build.

To make this happen, we have raised a $7M seed round led by Greylock, with participation from Amplify, Box Group, and SV Angel, and have backing from Reid Hoffman, Olivier Pomel and Alexis Lê-Quôc of Datadog, Ankur Goyal of Braintrust, Misha Laskin of Reflection AI, and current and former executives at Anthropic, Google Cloud, and Slack.

Interested in trying Niteshift? Try it yourself at niteshift.dev or book a demo. And if you would like to build with us, we are hiring.