The development of AI this year has been rapid. I believe many students have used AI to help themselves improve development efficiency, right?
For example, let AI generate code based on comments, explain the entire code, provide answers to technical questions, modify bugs, generate unit tests, etc.
At the just-concluded WAVE SUMMIT + Deep Learning Developer Conference on December 28, Baidu released a new product Comate AutoWork. This is a major release of the intelligent code assistant Comate. It has broken through the limitations and is no longer limited to the generation of code snippets and single files, but can help developers easily deal with complex projects and improve enterprise development efficiency.
So what’s the “magic” of this product? How to help us improve the development efficiency of complex projects?
What can evolved AI tools do?
Unlike existing AI coding tools, developers do not need to design solutions and write code step by step by themselves, but directly become "product managers" and make demands on AI. Next, the Comate AutoWork intelligent assistant will independently understand the requirements based on your needs => dismantle the work => formulate work plans and plans => become familiar with the project and generate code => debug and run.
In other words, AI no longer just generates code according to the developer's orders, it is more like your intelligent programming partner who can fight side by side with you and participate in the collaborative development of the entire project. This is not only an evolution of AI coding tools, but also an innovation in the human-machine collaborative R&D process!
Let’s take a look at Baidu’s official demonstration video.
, duration 01:45
Text saving:
1) First make a request to Comate AutoWork to add a button to an existing website project. When the user clicks, a carousel will pop up.
As shown below, you can directly give AI a Word version of the requirements document:
2) AI will autonomously understand your needs and share its thinking process with you. Just like a real person, it will read the entire project code; when it encounters requirements that it does not understand, it will proactively confirm with you.
3) After understanding the requirements, AI will break down the work and give its own execution plan. For example, first find the page to be modified, understand the old code, and even check whether there are components that can be directly reused!
We can view and modify the execution plan given by AI. If we find any errors in AI's understanding, we can manually correct them in time.
This step is really important. I feel deeply that when leading a team to develop complex programs, we must give them more guidance based on their ideas and plans to prevent them from developing according to wrong methods and steps.
4) After confirming the execution plan, AutoWork begins to work. We can have a cup of coffee and happily watch it get the job done step by step.
5) Finally, the AI did its job and wrote a bunch of code. We can read and adopt AI code one by one like Code Review without worrying that the code it writes will invade our project. It is clean and hygienic~
As shown in the video, Baidu Comate AutoWork finally completed the development of this "limited time feedback" task with certain complexity according to the requirements (including additional requirements) and Q&A/confirmation with engineers.
my thinking
I don’t know what you think after watching the above demonstration. At least I think this AI tool is of great significance to our R&D students and enterprises.
As programmers, our perspective in the past was that of "code farmers" and "brick builders". We had to do all the work ourselves, analyze the requirements ourselves, and write the plans and codes ourselves. It would be good if we could complete one requirement in a week. Now, using Comate AutoWork R&D Assistant, our perspective is that of "architect" and "manager", and more of our work is to propose requirements, guide AI, review plans, and review codes. Not only can it liberate us from boring work, but we can also learn a lot of new knowledge and R&D methods (such as dismantling work and debugging bugs) through interaction with AI.
For enterprises, if the Comate AutoWork tool is introduced, the developers of the team can focus more on innovation, release more productivity, and improve overall R&D efficiency.
Come and eat crabs
How to use the intelligent R&D assistant mentioned above?
First of all, Baidu Comate Intelligent R&D Assistant is open to all individual and enterprise users, and supports SaaS version and enterprise private customization.
If you want to try out AutoWork, you can go directly to the Baidu Comate official website to apply for a trial.
Directions: https://comate.baidu.com/
In addition, Comate has also released a new open plan . Enterprises can combine their own attributes, needs, and private domain knowledge to customize exclusive R&D plug-ins, fine-tune large models on demand, etc., to create their own intelligent R&D assistants.
This ability is really practical. For example, our company has developed some common development templates and components internally, which can be connected to Comate. Next time it generates code, it can think of reusing our existing code. For another example, each company has its own R&D specifications. You can connect the specification documents to Comate, and then the code it generates will follow the specifications.
Recommendations
If our company wants to use intelligent R&D assistants, the most important things we should focus on are accuracy, performance and safety.
Let’s talk about accuracy first, that is, whether the generated code is reliable and usable. Currently, according to official data, I know that Baidu Comate intelligent R&D assistant has more than 8,000 corporate partners, and the code adoption rate exceeds 40%, which is already a good result. As the AI large model continues to be enhanced, the assistant will become smarter and be able to provide code that better meets the requirements.
Then there's performance, which is how fast you can generate code. After my own trial, Baidu Comate's code generation performance is very strong and silky smooth.
When it comes to security, Comate does a pretty good job. The first is user privacy and security. The official promises not to store or analyze user code. When generating code, it will desensitize it before requesting it. It also supports SaaS and privatized deployment to ensure that data will not flow out. In terms of content security, Baidu conducts risk detection and filtering on model training data and inference content. In short, you can use it with confidence.
Many friends compare domestic AI R&D assistants to Microsoft's Copilot. In fact, I have used Copilot in the earliest days, and the accuracy of its generation is indeed very strong. However, after testing, Comate works out of the box and is superior to GitHub Copilot in terms of functional completeness, deployment flexibility, and performance. The most important thing is higher security. Based on this alone, domestic AI must stand up!
OK, I’ll share it here, and interested friends can give it a try. Students who have learned this, please give it a like or watch to support it, thank you!
