AI coding agents can write code, Crafting wants to help them ship it

AI coding agents are getting very good at generating code, but for enterprise engineering teams, that’s only part of the battle. There’s also testing, validation, and actually shipping that code against real production infrastructure — and that remains a bottleneck that most agent tooling doesn’t really address yet.

Crafting, a San Francisco startup founded by former engineering leaders from Google, Meta, Uber, and Discord, wants to fix that by giving engineers a platform that provides AI agents with production-like environments with real dependencies and real data to operate in and test their code.

Crafting CEO Sumeet Vaidya tells The New Stack that “about six to nine months ago, everybody was focused on faster code generation with AI agents. Yisui Hu, my co-founder, and I have seen what breaks when engineering organizations scale. And we felt that everybody was going to run into similar problems with agents at scale as well. General orchestration, coordination, efficient resource usage — all of that. ”

The company announced the general availability of Crafting for Agents on Monday, alongside a $5.5 million seed round led by Mischief.

From sandboxes to production-like environments

Crafting already provides cloud-based development environments for human engineers, with features like Kubernetes interception and hot-swappable services that let developers test against production-like setups. The new Crafting for Agents launch now extends that same infrastructure to AI agents.

The idea here is that a coding agent generates code the way it does today. But then, a separate testing agent spins up an environment with production-like dependencies via Crafting, where it runs its tests and iterates.

Crafting co-founders Yisui Hu (CTO) and Sumeet Vaidya (CEO) (credit: Crafting).

As Vaidya puts it, “Agents in sandboxes can’t really do anything. […] The way we see it, we’re providing agents with controlled access. This agent is trying to do this task, which means it needs access to our payments infrastructure at our staging tier. So let’s give them those credentials.”

Crafting’s bet is that for large enterprise customers, the hard part isn’t spinning up a container, it’s replicating the full complexity of a company’s infrastructure, from its network topology to its credential management to its compliance requirements.

“Before Crafting, every path we explored was either a point solution or something we had to stitch together ourselves,” says Cheuk-man Kong, Senior Engineering Manager at Faire. “Now, thanks to Crafting, we have scaled our entire agentic stack, enabling on-demand agents with access to internal systems, MCP servers, and cloud resources in a secure and scalable way.”

Credit: Crafting.

Enterprise onboarding

That also means Crafting’s onboarding process is inherently more hands-on. The company learns each customer’s network topology, configures Kubernetes clusters that mirror their production setup, and manages access to credentials. Early customers include Brex, Faire, Webflow, Verkada, Persona, and Instabase.

Vaidya argues that having internal staging environments that mirror production doesn’t really work. So instead, the team helps its customers set up environments and spins up a cluster that “directly mimics as much as what they need.”

Vaidya notes that every single enterprise is “very different in terms of their infrastructure and their expectations. There will always be some bespoke white-glove service that we offer to make sure things are up and running.”

He says most of the interest and the fastest adoption have been in fintech and heavily regulated industries, where companies have very clear requirements for security and controlled access. And it’s not just those tech-first firms that need help. ”Outside of Silicon Valley, companies are learning the same lessons a lot of us grappled with a year ago,” he says.

The company says teams using Crafting are shipping 25 percent more pull requests quarter over quarter, with engineers saving about 2.5 hours a week on environment setup alone. And across its customer base, AI-generated code has scaled from single-digit percentages to as high as 70 percent of total output within twelve months.

“The developer story is great, but it’s becoming very clear that the agent story is not just important, but actually a critical factor for all these companies now too,” he says.

The bigger vision

Vaidya, who spent four years as a director of engineering at Discord before founding Crafting (and held roles at Uber and Facebook/Meta before that), sees software development as only the starting point.

His longer-term vision is to turn Crafting into what he calls an “operating system for agents” — a general-purpose infrastructure layer where any enterprise agent can acquire the credentials and access it needs, when it needs them, whether that’s for observability, monitoring, or collaboration.

For now, though, the focus is on building out the engineering team and expanding internationally. The broader question for Crafting — and for the agent infrastructure market in general — is whether enterprises are ready to invest now, given how fast the underlying technology is changing.

And Vaidya argues that there is no point in waiting.

“You could say making an investment right now, it won’t be as good as if you built it in six months,” he says. “But you would have six months of productivity and positive output from what you built today.”

The post AI coding agents can write code, Crafting wants to help them ship it appeared first on The New Stack.