Meet Devin, the first AI software developer

A Cognition AI, a startup recently backed by American billionaire Peter Thiel (co-founder of PayPal), announced this week the launch of “the world’s first AI software engineer”. Named after Devin, this innovation promises not only to offer suggestions, but also to autonomously develop and execute complete digital products, carrying out testing, bug review and all stages of a project. According to the startup, Devin is able to solve 13,86% of the tasks in the SWE-bench without human assistance, an efficiency test where the highest result achieved by an AI (Claudia 2) was 4,80% until then.

A Devin represents a significant leap in the field of artificial intelligence, showing not only the ability to suggest, but to autonomously create complex digital products. Its autonomous development and execution of websites, applications and games redefines the way software development is considered. However, the impact of this innovation goes beyond technological advancement, also representing a challenge for professionals in the field who will need to retrain themselves in the face of the threat of automation of previously exclusively human tasks.

How Devin works

A Devin is an AI model software engineer who serves as a tireless teammate as he works without time limits, ready to build software alongside you or complete tasks independently for later review. Its ability to plan and execute complex programming tasks, involving thousands of decisions, is impressive, being able to remember the relevant context at each step, learn over time and correct development errors.

A Devin was equipped with common developer tools such as shell, code editor and browser, all within a sandboxed computing environment. One of the most interesting features of Devin is your ability to actively collaborate with the user. It reports its progress in real time, accepts feedback, and works with the user on design choices as needed.

A Devin It works in a simple and intuitive way. The human user begins by interacting with the AI ​​chatbot-style interface by typing a natural language prompt. From this prompt, the Devin gets to work, developing a detailed, step-by-step plan to solve the problem at hand. If something doesn't feel right during the process, the user can enter the chat interface and give the AI ​​a command to correct it. This ability for direct interaction and instant correction allows engineering teams to delegate projects to Devin and focus on more creative tasks that require human intelligence.

Although other AI tools for developers — like Microsoft's GitHub Copilot — offer assistance features for programmers, such as automatically completing chunks of code and translating codes between multiple coding languages, they cannot complete codes end-to-end without human interference or assistance — which is the big differentiator of Devin.

How to use Devin

A Devin AI is currently available with restricted early access for those interested in using the AI ​​agent for engineering work. Customers can request early access to AI through company website, since to date, the Devin it remains non-public, with the company only granting access to a select few customers.

A Cognition AI has released several videos showing developers using AI to develop new products in an impressive way. These videos highlight the power of AI in action, showing how it can streamline and optimize the software development process. The capacity of Devin is put to the test to create and deploy end-to-end applications, as in the case where we made an interactive website that simulates the Game of Life, gradually adding features requested by the user and then deploying the application to the netlify.

Another essential function of Devin is to find and fix bugs in codebases autonomously. For example, in this video below she helped Andrew maintain and debug his open source competitive programming book. A Devin It's also capable of training and tuning your own AI models, like setting up fine-tuning for a large language model with just a link to a research repository on GitHub.

Additionally, Devin can troubleshoot bugs and feature requests in open source repositories, contribute to mature production repositories, and even perform real-world tasks like writing and debugging code to run a computer vision model. These skills of Devin represent a significant advance in the field of artificial intelligence applied to engineering and software development.

In the video above, also released by Cognition AI, Devin was able to resolve a bug with logarithm calculations in the algebra system Python sympy impressively. The AI ​​set up the code environment, reproduced the bug, autonomously coded and tested the fix. This example demonstrates the ability of Devin to identify and solve complex code problems, demonstrating your ability to deal with programming challenges efficiently and accurately.

Is this the end of programmers?

A Devin, heralded as the world's first software engineering AI, is at the center of an intense debate about the future of human programmers. Its launch caused quite a stir on the web, with many questioning whether this technology could eventually replace programming professionals. However, even with the impressive capacity of Devin to solve 13,86% of the tasks in the SWE-bench Without human assistance, experts agree that it still cannot completely replace a human programmer.

The fact that AI is capable of performing complex software engineering tasks raises questions about the role of programmers in the future. Although the Devin While AI can be effective in many areas, such as writing code, fixing bugs, and even learning to use new technologies, there are certain human skills, such as creativity, intuition, and critical thinking, that are difficult to replicate in an AI. These skills are essential for solving complex problems and creating innovative solutions, something human programmers bring to the field.

Furthermore, AI like Devin offers significant advantages in terms of efficiency and availability. While a human programmer is limited to a working schedule and the ability to focus on one task at a time, AI can work 24/7 and handle multiple tasks simultaneously. This can result in greater productivity and speed in software development.

Devin is the most impressive display I've seen in the last decade. It's worth a few minutes out of your day to watch and get a glimpse of the future, it's surreal and deeply moving.

Eric Glyman, co-founder of startup Ramp.

However, despite these advantages, AI is unlikely to completely replace human programmers. Instead, it's more likely to be used as a complementary tool, helping programmers be more efficient and handle low-level, repetitive tasks while humans focus on more complex and creative tasks. Therefore, rather than spelling the end of human programmers, the Devin and other similar AI represent, at least for now, an evolution in the way programming work is done, with humans and AI working together to create innovative and effective solutions.

See also:

Sources: Cognition AI, Business Standard e VentureBeat

reviewed by Glaucon Vital in 15 / 3 / 24.