Infrastructure automation company Pulumi have introduced what’s claimed to be the first artificial intelligence-based platform engineering agent for the industry, named Neo. The tool works to resolve some of the infrastructure bottlenecks that develop as a side effect of AI tools speeding up software development. Neo was launched in public preview on September 16th as Pulumi’s first AI-powered platform engineering agent.
In a press release, Pulumi explains how the speed of platform engineering teams has become challenging to sustain due to the advances AI coding assistants have made in speeding up software development. According to Pulumi’s Founder and CEO, Joe Duffy, AI has completely transformed application development methods over the past twelve months. The company has identified a “velocity trap”, which happens when AI tools enhance developer efficiency, but platform teams fail to maintain the same pace.
The AI-powered Pulumi Neo operates as a built-in agent within Pulumi Cloud, performing automated infrastructure work while maintaining enterprise-level governance. It is a fully agentic platform engineering AI agent that automates infrastructure provisioning, management, and optimisation end-to-end. It has built-in governance, compliance, and multi-cloud support.
Neo can independently execute complex tasks, understand dependencies across cloud resources, and generate detailed previews and histories for all automated actions. Other key capabilities include fully agentic workflows with approval processes, deep integration with infrastructure as code (IaC), multi-cloud context awareness, and automatic enterprise guardrails that respect existing governance settings.
Writing in a blog post, the Pulumi Neo Team explain the reasoning behind the tool:
Platform engineers don’t need faster AI tools. They need their own dedicated AI tool, one that understands infrastructure context, speaks their language, and works within their constraints.
According to the press release, the initial beta users of the tool saw notable improvements in their infrastructure management processes. Werner Enterprises reduced infrastructure provisioning time from three days to four hours, allowing their development teams to deploy features at 75% faster rates while maintaining SOC 2 compliance. The press release quotes Jason Harris, Cloud Architecture Lead at Werner Enterprises, who notes, “We have a high tolerance for AI capabilities and we’re actively looking for where we can integrate AI without compromising our infrastructure governance and reliability requirements.”
According to Pulumi, Neo becomes increasingly capable as organisations invest more deeply in infrastructure as code practices, and the blog post introduces the concept of a “reinforcement cycle”. Every Pulumi component written, policy defined, and environment configuration established becomes part of Neo’s operational context, informing its decision-making processes. The system is designed to enhance human expertise instead of replacing it, enabling engineers who grasp infrastructure reasoning to become more valuable through the agent’s ability to execute large-scale implementation tasks. Pulumi claims that teams which make best use of the reinforcement cycle are more proactive with policy creation and system design, compared to others who must stay reactive by responding to issues and manually updating versions. They claim that the accumulated organisational knowledge directly enhances the AI’s effectiveness.
Pulumi published a technical demonstration on their YouTube channel which displayed Neo’s capability to execute complex yet regular operations including AWS runtime updates. The system processed an AWS end-of-life notification email to identify outdated Node.js runtimes across multiple repositories before recommending an update to the current LTS version. It also established a new policy to prevent similar future occurrences. The agent also automatically ran Pulumi previews to validate changes before implementation, ensuring that no resources would be erroneously destroyed or recreated, while also maintaining policy compliance.
The system helps large organisations solve platform team scalability issues that arise during infrastructure management. The YouTube demonstration explains that the average Pulumi customer operates more than 20 infrastructure-as-code repositories, while the largest customer has over 230 repositories.. Any infrastructure modification at this scale becomes complicated because organisations need to understand both the necessary changes and their organisational effects.
Similar agentic assistants are also available. GitHub Copilot for Infrastructure focuses primarily on AI-powered code completion, real-time suggestions, and documentation generation for infrastructure as code (IaC) in editors like VS Code. Its strength lies in generating Terraform, Pulumi, or other IaC scripts, providing boilerplate, refactoring code, auto-documentation, and converting comments to working infrastructure code using advanced language models.
AI DevOps Platform company Harness has also developed an AI solution called the AI DevOps Assistant which takes a different approach to DevOps problems. The Harness tool focuses on pipeline development and management through the ability to create complete pipelines and workflows using natural language commands.
The Harness assistant focuses on pipeline optimisation through automated best practice recommendations and failure detection, shortening the time needed to recover from errors. This tool focuses on pipeline management, whereas Neo claims to handle infrastructure management across the board.
By using or expanding existing infrastructure-as-code, policies, configuration, it gets smarter over time. It’s built to work with what teams already use.
– Dinesh Sharma on LinkedIn
The release of Neo is Pulumi’s attempt to address the emerging problem of how platform engineering teams can operate successfully in an AI-driven development environment, offering specialised tooling explicitly designed for infrastructure challenges rather than generic coding assistance.
