InsForge has completed an $8 million seed round and launched InstaCloud, a cloud platform built for AI agents. The AI infrastructure startup was co-founded by Hang Huang and Tony (Yao-Wen) Chang, and is now part of Y Combinator Spring 2026 (P26). It is based in San Francisco and has a team of about six people.
Targeting the infrastructure bottleneck after code generation
ABMedia said it had previously covered InsForge in May, focusing on how the bottleneck in software development has been shifting from writing code to databases, backend systems, deployment, and cloud infrastructure after the rise of AI coding tools. The report noted that coding agents such as Cursor and Claude Code can already generate large amounts of code, but what often stops AI from independently completing a product is not coding itself. It is the infrastructure work that comes after.
That is the problem InsForge is trying to solve. The company frames it as a response to the question: what happens after AI finishes writing the code? Its approach is to design the entire backend as an interface that AI agents can operate directly. YC describes InstaCloud as an "agent-native serverless cloud," built so AI agents can deploy applications and services, scale them automatically, and minimize the need for human involvement in DevOps.
YC’s latest introduction also describes the product as infrastructure that allows coding agents to manage the backend from start to finish. In the report, agents can use a unified CLI to operate services including servers, databases, authentication, and storage, instead of moving across several dashboards and manually copying settings the way human engineers often do.
More than 50,000 projects have been created
According to a recent report by n startup, InsForge has completed the $8 million seed financing. The company launched InstaCloud at the same time, aiming to meet new cloud demand created by the growing role of AI agents in software development.
Hang Huang said developers have already created more than 50,000 projects through coding agents on the platform. He also said InsForge’s monthly GitHub commit activity doubled within four months, with the amount of code actually executed and deployed rising quickly.
Tony Chang said AI agents are building software at a scale that traditional cloud systems were never originally designed to handle.
The product idea was shaped in part by Tony Chang’s Databricks experience
The report said part of InsForge’s product concept came from Tony Chang’s experience as a Network Infrastructure Engineer at Databricks. YC’s official materials also list his past internships at Meta Ads ML Infra and Amazon Backend Infra.
Chang had once been responsible for investigating a backend deployment issue. He spent two to three days tracing the problem through the infrastructure, but the actual fix took only about 10 minutes. That experience led the team to ask why engineers spend days locating a problem if the real repair can be so simple.
In their view, backend issues are often spread across databases, permissions, servers, and integrations between different services. The hard part is understanding the state of the whole system. That, they believe, is well suited to agents.
The report said that if AI can directly read infrastructure state, runtime logs, and configuration, then call tools on its own to fix issues, its role is no longer limited to helping write code. It starts to look more like an engineer that can operate in a production environment.
Rejected by YC five times before being invited in
InsForge had previously been rejected by Y Combinator five times. According to the report, the company at that stage looked more like a traditional backend developer tool and had not yet shown how it could stand apart in a highly competitive market that included Supabase, AWS, and Railway.
The team kept shipping products and adding users. After its first large-scale launch, InsForge recorded more than 1.5 million impressions on X, reached No. 1 on Product Hunt, and appeared on GitHub Trending. Its user base later grew to about 30,000, including close to 300 paying customers.
In the end, the team did not return with another application. YC saw the product’s growth, contacted InsForge, and invited the startup to join the Spring 2026 batch.