Over the past few years, LLMs have evolved from sophisticated code completion tools to AI agents that can create software, construct and fix entire modules, and increase the productivity of software engineers.

Some of the excitement surrounding AI-powered software engineering agents is unfounded hype, just like many other things involving LLMs. However, developers who master the new generation of AI technologies will be able to accomplish much more in less time, and there is also genuine value to be gained.

Developers can use some of these five interesting agents to automate monotonous chores like writing documentation, cleaning bugs, automating code, and handling pull requests.

Cody.AI

Cody is a customizable AI helper for your company. Utilize Cody to assist your team, respond to inquiries, resolve problems, and generate concepts. Cody solves your business questions promptly, saving you the trouble of reviewing documentation. You can upload various data formats, including PDFs and PowerPoints, to build a customised knowledge base. Cody will use this knowledge to formulate thoughtful answers and consistently cite sources to establish his credibility. 

Writing emails, translating documents, and producing marketing materials are just a few of the things it can assist with. Cody also helps troubleshoot support documentation difficulties and can provide insightful suggestions and guidance based on contextual history.

Grit.io

The goal of the developer tool Grit is to make program maintenance easier. It includes an optional CLI for local management and a web interface that allows the creation of pull requests through automated end-to-end migrations. Fundamentally, Grit uses two main tools. GritQL, a powerful and intuitive query language that makes code alteration through static analysis easier, is the first. With the help of AI-powered transformations, the second tool enables migrations to conform to the rules of your codebase smoothly.

When combined, GritQL and AI transformations effectively manage tedious duties related to updating out-of-date code, freeing up your time customising the program.

ReactAgent

Based on the GPT-4 language paradigm, ReactAgent is an autonomous agent at the experimental stage that creates and arranges React components according to user stories. This agent's technology stack consists of Radix UI, Shandcn UI, Typescript, TailwindCSS, React, and OpenAI API. The agent uses Typescript, TailwindCSS, and RadixUI to construct pertinent screens after interpreting user story content and effectively generating various React components using atomic design principles. 

Even though ReactAgent is still in the experimental stage, the first results are intriguing and encouraging. The project welcomes community contributions and is completely open-sourced. ReactAgent's ability to create React components from user stories, create React components using pre-existing components, leverage a local design system for React component generation, and follow Atomic Design Principles are some of its noteworthy characteristics.

Bloop.ai

Bloop is a code search engine that combines GPT-4 and semantic search to enable effective searching. Users can use a natural language approach made possible by GPT-4 or use more conventional techniques like literals or regex to search their proprietary codebases. The technology facilitates accurate code navigation, allowing users to navigate between definitions and references within the code. Bloop employs neural semantic code search by comparing encoded meanings in vector representations of queries and code snippets, in contrast to conventional code search tools focusing on term matching. 

Bloop maximizes code relevance and answer accuracy by integrating a semantic search engine and using GPT-4 to generate keyword queries. Users have the option to narrow down their searches by programming languages or repositories. Bloop's open-source software has multiple features in one repository that anyone can use. The software is a free desktop application that uses MiniLM embedding models to index locally, protecting codebase privacy.

BitBuilder

With the skills of a novice software engineer, BitBuilder serves as a virtual coding helper. After understanding the explicit criteria, the agent creates Pull Requests straight from your repository. Users can work on the branch BitBuilder starts or contribute to speed up development. The efficient production of Pull Requests is one of BitBuilder's main use cases. Users can examine a Pull Request and create a GitHub problem. 

Within two to five minutes, BitBuilder creates a PR and quickly develops an implementation strategy. This feature promotes using BitBuilder's first attempt at a code update. BitBuilder also makes it easier to respond to code comments. Users can work with BitBuilder without reading the code to manage code review comments straight through GitHub. This capability increases the development workflow's efficiency.

Image source: Unsplash