Augment has developed a new AI agent designed to address one of software development’s toughest challenges: working with large, complex codebases. This vibecoding-supportive chatbot supports memory and model context protocol (MCP) and can handle up to 200,000 context tokens.
“Until now, no AI coding platform has specifically targeted the challenges software engineering teams face working with large complicated codebases,” Augment co-founder Guy Gur-Ari tells me in an email. “Augment is uniquely focused on building AI for professional software engineers. Our team has deep expertise in AI, as recently proven by our top spot on SWE-bench-verified.“
The Augment Agent is a step up from the company’s October 2024 offering, which competed against GitHub Copilot, Cognition’s Devin, Codeium, Blackbox AI, Amazon Q, and other AI-powered coding assistants. Gur-Ari points out that although GitHub Copilot “paved the way for developers using AI in their workflow,” it loses its luster when dealing with large codebases.
“Copilot is like pairing with a brand-new developer: They may be smart and fast, but they don’t know enough about your codebase or context to be useful,” he states. “Augment is an expert in your codebase and provides the most relevant and useful suggestions, where you work, at lightning speed.” Compared to GitHub’s Copilot, he boasts that Augment has a 70 percent win rate in deals.
Making Augment Agent Work in the Right Context
To ensure that its AI is generating the correct code in the context of the developer’s environment, Augment has infused its agent with Memories, an automatically updated feature that persists across conversations to improve code generation continuously, complete tasks faster, and match their code style and pattern. The company has also integrated Model Context Protocol (MCP), a process standardizing how applications provide context to Large Language Models, into Augment Agent.
“MCP allows you to connect to other tools and systems to bring in more relevant information to your work,” Gur-Ari explains. “It’s like working with someone who has been at your company for years and knows everything there is to know versus someone who is very smart but has limited understanding of your business. The more data you can bring in, the better the AI-driven result.”
In addition, the company has developed native tools that enable the agent to work with popular third-party applications. It initially launched with five integrations: GitHub, Jira, Confluence, Notion, and Linear.
Another key improvement to Augment Agent’s coding effectiveness is its expanded context capacity. The agent supports up to 200,000 tokens, a design choice specifically made to accommodate large and complex codebases.
With smaller context windows, important information can be cut off—leading to incomplete understanding and a higher risk of hallucinations. In contrast, Augment’s larger capacity gives the agent a broader view. It can process entire monorepos, pull in detailed context from integrated tools like GitHub, Jira, and Notion, and maintain consistency in style, structure, and logic across its code generations.
“Expanded context is a competitive advantage for Augment,” Gur-Ari asserts. “Our focus is large, professional software organizations and production-grade applications. The large context window allows Augment Agent to process and understand large, complex codebases that would overwhelm competitors with smaller context windows.”
Subscribe to The AI Economy
How Much Does Augment Agent Cost?
Augment acknowledges that pricing for AI agents can be complicated. That’s largely due to the high cost of compute, model access, and the lack of reliable data on how frequently developers use these tools.
To address this, the company is introducing an unlimited Early Adopters plan. This approach is designed to reduce friction for new users and avoid potential pushback as the product rolls out.
The pricing model follows a freemium structure. It starts with a Community Plan, which is free to use. However, usage is capped—developers can make up to 50 agent requests per month. If they need more, additional requests cost $9 per 100.
For more active users, there’s a Developer Plan priced at $30 per month. At launch, this tier includes unlimited agent requests. However, Augment notes that over time, this will change: eventually, users will be able to purchase additional requests at a rate of $11 per 100.
Lastly, there’s an enterprise offering with unlimited agent requests and additional features such as Slack integration, team management features, usage analytics, SOC 2 Type II reports, and custom terms.
The Next AI Step For Enterprise Development Teams
Founded in 2022 by Scott Dietzen, the former CEO of Pure Storage; Igor Ostrovsky, ex-Chief Architect at Pure Storage; and Gur-Ari, a former Google engineer, Augment Code is pursuing developing an ideal AI code assistant for the enterprise. “I continue to be struck by how that joy of programming solo I experienced fails to translate to far more talented teams building far more meaningful software,” Dietzen wrote back in August. He outlined some of the issues plaguing software teams:
- Too much drudgery sorting mundane code and tests;
- Too much time trying to understand code;
- Too much toil in updating libraries and dependencies;
- Too much effort required for code reuse and refactoring;
- Too much time doing code reviews and ramping newbies; and
- Too many meetings trying to coordinate efforts.
Enterprise codebases are large, complex, and deeply interconnected, built on years of legacy code, cross-functional dependencies, and extensive documentation. It could be argued that many AI coding assistants aren’t capable of handling these situations since they don’t have that large of a context window or can’t comprehend the nuances of enterprise-grade systems.
Dietzen once wrote that GitHub Copilot was the equivalent of a fresh computer science graduate instead of a programmer steeped in an organization’s environment. “While they indeed look over the shoulder of an individual programmer, they are unaware that the programmer is part of a broader team; and while they make it easy to add code, they do not sufficiently help reuse code, so your software gets more bloated, more complex, and more fragile.”
Augment Agent is available today for VS Code and JetBrains through a 14-day free trial. Gur-Ari states the company is focused on bringing its tools to “wherever professional developers do their work” and is looking at “expanding to other platforms in the future based on user demand.”
Featured Image: Augment
Leave a Reply
You must be logged in to post a comment.