Is AI Officially Taking Our Jobs? Meet Devin, The First AI Software Engineer

Here's why everyone is talking about Devin AI.

Created with Midjourney

Cognition AI just unveiled Devin, branded as the "first" AI software engineer. This groundbreaking AI was developed in a mere five months, rapidly securing millions in funding within days.

Upon hearing about this latest launch, the immediate question that springs to mind is whether Devin might be the AI tool that finally makes human programmers obsolete, or if it's merely another fleeting AI craze that will fade from memory in a few months.

But before we tackle that question, it's worth taking a closer look at why this so-called “AI software engineer” has captured so much attention recently.

Devin isn't the first AI to target this niche and boast such capabilities. There's GPT Pilot (also known as Pythagora), Meta GPT, and several other lesser-known projects that can arguably match or even surpass Devin's performance. So, what sets Devin apart?

Here are three reasons why they were able to achieve such an explosive launch:

1. On day one, they secured a hefty sum of money, totaling nearly $21 million in Series-A financing to date, headed by the Founders Fund (among the most renowned and respected entities in Silicon Valley, initiated by Peter Thiel, PayPal's founder and an early investor in Facebook).

2. Scott Wu, the brains behind Devin, is a math genius who began programming at age 9 and later won gold medals at the International Coding Olympiads. It's hard to overlook him, especially when you consider that the Cognition Labs team consists of 10 programmers who have also proven their skills in international programming competitions. As a result, Wu and his team have been hailed as a collective of geniuses. Their work on developing an AI has captivated thousands of investors, users, and members of the scientific community.

3. Although I said before that Devin isn't technically the first AI engineer, the fact that Cognition introduced it as the first AI software engineer in its YouTube video led many to accept this claim as true.

Getting to know Devin

One capability that sets Devin apart is its ability to plan and execute complex tasks, saving time in decision-making processes to achieve desired outcomes. This feature, however, isn’t unique to Devin; other AIs available in the market are equipped with similar functionalities.

So, what makes Devin different?

The answer lies in its user interface. It incorporates tools that developers typically use in their day-to-day tasks, including a shell, code editor, and a browser within an isolated computing environment, alongside a chat function that facilitates ongoing interactions with the AI.

Devin Interface

It's genuinely remarkable because you'd usually need to switch between various work environments when working with a coding assistant. This gives Devin a significant advantage.

Additionally, I want to highlight some practical use cases where Devin

• Can learn how to use unfamiliar technologies

• Build and deploy apps end to end

• Autonomously find and fix bugs in codebases

• Train and fine tune its own AI models

• Address bugs and feature requests in open source repositories

• Contribute to mature production repositories

• Complete actual assignments on Upwork

In general, the demonstrations of Devin's power are astounding. However, at this point, I must say that since Devin is not yet available to everyone, these videos could merely be a well-edited showcase of the best responses (somewhat similar to what Google did with Gemini not so long ago).

Also, this AI isn't open source, meaning you can't integrate it with your own models for local execution.

Nonetheless, an important point to highlight is Devin's ability to access GitHub repositories and carry out tasks such as analyzing, extracting, correcting, and building on vast amounts of code, a feature many current AIs lack.

Regarding performance, the image below suggests that Devin significantly outperforms other AIs.

Devin Performance

But is this a good comparison?

We might argue that the comparison isn’t completely fair. This is because Devin is an agent, which implies that it likely has other agents working together to perform a specific task. This means that tasks such as iterating over lines of code until the correct solution is found are considered as a single task or as being accomplished on the first attempt (0-shot). This makes the comparison with models like Claude 2 and GPT-4 not entirely appropriate.

Ideally, a comparison with other agents like Pythagora, GPT Pilot, Meta GPT, or Super AGI would have been more convincing.

More than just a simple coding assistant

Devin is undeniably impressive. Over time, many things as we know them today will change and adapt, and programming is no exception. However, this project wasn't primarily created to replace the tasks of a programmer. The approach behind its design encompasses something deeper: the goal with Devin is to enable AI to reason.

Does this sound familiar? It might remind you of the goals associated with AGI (Artificial General Intelligence)

But when it comes to reasoning, robotics begins to play a crucial role, and that's what we often overlook. That human action (which for us is the most normal thing) represents one of the biggest challenges for AI. While Devin's performance in software engineering seems spectacular (SWE-bench, 13.86%), it's still far from that 100% mark, which would signify the construction of software without human supervision or assistance.

Achieving that ideal percentage signifies a shift in how AI models are currently being trained. To achieve 100% reasoning (or anything close to AGI), new systems must be created that can interact with the physical world.

In other words, robots.

Tech industry giants and AI researchers understand that to transcend the current paradigm of model training, we need to look beyond readily available data, which is a resource poised to diminish over time. This implies that human-generated data will no longer be enough.

As it stands, AI systems are often quick and efficient, yet they lack the creative essence uniquely attributed to humans. To embody such a human trait requires a profound comprehension that AI, if left to its own ways, would struggle to achieve.

I remain optimistic that AI will eventually develop reasoning that surpasses human capabilities, which, in the long run, will prove beneficial for humanity. Time will tell whether Cognition will achieve its goal with Devin.