Artificial intelligence-enabled software engineering comes at a cost. Yes, AI tools can write more code – and do so more quickly – than human engineers. But that code often carries more errors and even where it doesn’t, the sheer volume of code inevitably means more post-production incidents to respond to and more security vulnerabilities to remediate. This is causing real stress: Gartner predicts 40% of AI-augmented coding projects will be cancelled by 2027 due to escalating costs, weak risk controls and other issues.

Autoheal thinks it can help. The San Francisco-based start-up, which is today announcing it has raised $7.9 million of new funding, describes its innovation as a “self-improving software factory”. In a nutshell, Autoheal’s AI agents take on much of the post-coding work for enterprises, promising to respond more quickly to problems as they emerge with solutions that human engineers then decide whether to implement. The platform also evaluates the performance of these agents, with the aim of improving their future responses.

“Every completed task should help the company do better next time,” explains Sid Choudhury, co-founder and CEO of Autoheal. “Once an engineer gives a correction to the agent, we make sure that this learning becomes part of its memory so that next time, when a different engineer is working with the same agent, he or she is not being asked the same questions all over again.”

In a world where AI coding is rapidly becoming the norm rather than the exception, this becomes ever more valuable. Research from Astute Analytica estimates the AI coding assistant market will grow 10-fold between 2025 and 2035, to be worth more than $40 billion. But analysis from Sonar shows that while as much as 42% of code is now AI-assisted in some way, more than nine in 10 developers don’t fully trust the output.

Engineers therefore want more control over AI coding, argues Choudhury, but they don’t want to spend an ever-increasing amount of time going through AI coding assistants’ work to identify issues. “Autoheal connects to their systems and we investigate the problem,” he explains. “We highlight the likely cause so that engineers have something to work on and more time to act.”

While Autoheal was only launched last year, the business’s sales pitch appears to be resonating. It has already signed up blue-chip customers including the investment banking group Nomura and the accounting software provider AvidXchange.

Sameer Jain, CIO, Wholesale at Nomura Bank, says the Autoheal platform has enable it to cut the average time to resolution from two hours to 15 minutes in the case of many problem areas. “Our production operations teams spend valuable time triaging alerts and managing incidents, while also pulling engineers away from their software development activities,” says Jain of the problem Nomura faces. “The fact that [Autoheal] runs entirely within our own cloud, in compliance with our controls, made it a natural fit for how we operate.”

It’s an important point. One key to Autoheal’s early success, Choudhury says, was its recognition that customers would want to retain control over their data and IT assets. “We typically run behind their firewall inside their cloud systems, accessing their pre-approved LLM providers,” he explains.

Autoheal’s investors think the company could have a significant future as AI coding becomes ubiquitous. Today’s seed funding round is led by Innovation Endeavors, with participation from Emergent Ventures, U&I Ventures, Darkmode Ventures, Batch Ventures and Param Hansa Values.

“Enterprises are moving quickly from experimenting with AI agents to asking how they can operate them safely and efficiently at scale across the entire software factory,” says Innovation’s Harpinder Singh. “Autoheal is building the agent infrastructure layer that makes that possible.”