Augment AI Code Assistant Targets Large Development Teams

Software engineering is not a solo sport but most AI coding assistants are designed as though developers work alone. A new AI coding assistant released today, Augment AI, is designed for large software engineering teams, with support for collaboration and a developer manager dashboard that provides insight into the development lifecycle.

“Software development is almost universally a team sport,” Scott Dietzen, CEO of Augment AI, told The New Stack. “As you get to these larger teams and more complex code bases, the current AIs treat every developer as if they’re the only one in the code base, and that’s not at all close to reality.”

How Augment Differs From Individual Copilots

While other copilots offer an understanding of programming languages and algorithms, they do not have a deep understanding of a particular organization’s code, said Dietzen, who is himself a former Cobalt programmer.

“This is one of the core innovations we’re hanging our hat on, is that we are able […] to deliver that deep knowledge of customer code,” he said. “How work gets done, architectural conventions, security policies and best practices are all reflected in our AI.”

It does this by using Retrieval-Augmented Generation (RAG) to learn an organization’s particular code and practices. When organizations first start with Augment, they go through an indexing interval where the code is taken up to its public cloud instance, which is SOC II compliant, explained Dietzen.

“Instead of fine-tuning, we use retrieval augmented generation, and we have a real-time RAG implementation,” he added.

It’s a bit of secret sauce for how it selects the appropriate subsets of the code, however Dietzen did say the AI has a real-time view of what’s actually going on with developers, to ensure it’s up-to-date with code and practices. Also, Augment uses a combination of open source and closed large language models (LLMs).

“A typical user will be interacting with at least five different large language models of quite varying sizes for different kinds of use cases,” Dietzen said. “On the open source side, there’s always a lot of post-training we’re doing on top of them, but there are still things that the larger models do very well, and so we’re just trying to optimize for the best possible user experience by putting the right model to work at the right time.”

It currently supports more than 80 languages, including TypeScript, JavaScript and popular JavaScript frameworks, including React.

Customers have used the tool to migrate from C++ to Rust, move from one cloud service to another, or move from a NoSQL to a SQL database, Dietzen said.

Augment AI Integrates With VSCode, JetBrains IDEs

Augment AI is offered as a cloud service, with plugins for VSCode and the JetBrains family of IDEs, but it also integrates with collaboration tools such as Slack, he said. It can handle large repos, access privileges and frequent code changes, he added.

The Slack bot allows teams to have discussions and work out problems with code within Slack. For example, Dietzen shared an instance where a software engineer started a discussion in Slack about receiving an error message that didn’t seem to make sense.

“Another engineer said, ‘I don’t know, ask Augment,’ and then Augment came back and said, ‘Yes, this looks like this error message was delivered. It wasn’t the right outcome for this error message to show up in this use case. And here’s a proposed fix,’” he said. “Then one of the other engineers assigned a third engineer to go off and fix it. You were able to handle this whole thing inside of Slack and leverage all of the history and information that’s in Slack as well.”

AI for Development Managers

It also incorporates an engineering manager console to provide visibility for engineering leaders to use AI to understand how a team is doing and where there are issues. Managers can track across teams who are using AI and who isn’t, as well as information about how it’s being used and what’s working and what isn’t.

“For example, if they’re seeing greater productivity out of one team versus another team, they can check and see, is Augment being used in the same way between those two different teams,” Dietzen said.

“If I want to get the AI’s comments on a code change, it’s very easy for me to pass that along to whoever my code reviewer is … and maybe make their job a little easier.”
– Scott Dietzen, CEO of Augment AI

Augment can also help with changes within the development process, he added. For instance, it can notice if you’re changing a security model or changing a target database.

“We can actually spot those trends and attempt to accelerate them, so we become a natural vehicle to help push code along in the direction that leadership wants to take it,” he said. “We’ll even encourage developers to move toward a new module if it’s replacing an old module, for example.”

Another feature that Augment offers is easy sharing of the AI’s input through the IDE with different developers.

“If I want to get the AI’s comments on a code change, it’s very easy for me to pass that along to whoever my code reviewer is … and maybe make their job a little easier,” he said.

AI for the Frontend Developers

Customers are using Augment AI on the frontend as well as the backend. The company shared some real-world use cases customers have shared with them.

For instance, one frontend developer said her “ah-ha” moment came while trying to write unit tests. What surprised her is that it was able to make correct inferences even though the information wasn’t given in her view of the code.

The developer said the UI had actions that end up producing a custom query language, and Augment was able to handle code completion for a new type of action in that custom language, which is not something the team had been able to do with GitHub Copilot.

Another frontend developer was able to use the AI to understand Go code provided from a backend engineer. The developer simply asked the AI to comment on what the code is doing, which allowed the developer to easily create a JavaScript version without being able to code in Go.

Using AI to Reduce Onboarding of New Developers

Dietzen shared an example of how it can boost productivity. Webflow’s chief architect used to spend a lot of time answering questions about the code base, particularly for new hires or when developers moved from one project to another inside the company.

“Then he realized that Augment could do a better job of onboarding these new hires than even his senior engineering leads, and it didn’t take time away from him and his peers to have to spend a lot of energy ramping up these new hires,” Dietzen said.

That’s because Augment will summarize the current repository, providing the developer with an overview of its purpose. Developers can also query Augment, receiving feedback in real-time without bothering anyone else on the team, he added.

It can also answer questions about why an application is doing something, like using two different databases or what the organization’s security protocols are, making it easier to shift left during code creation.

“Because Augment has this deep knowledge of customer code and how software gets built inside of the environment, all of the questions that typically prevent new engineers from being productive, that they have to amass this knowledge of both how the code base works, where everything is and how to get things done,” he said. “They now have an AI working with them that can give them all of those insights.”

The post Augment AI Code Assistant Targets Large Development Teams appeared first on The New Stack.

A new code assistant platform offers AI insights with support for what large software development teams need, such as collaboration support.