But a competitive landscape for coding platforms supported by AI is crowded. Windsurf, replit and pasolside starts also sell AI generating tools for programmers. Cline is a popular open source alternative. Github’s Copilot, which was developed in cooperation with OpenAI, is described as a “couple programmer” who codes automatic supplement and offers debugging aid.
Most of these code editors are to combine AI models built by main technology companies, including OpenAI, Google and Anthropic. For example, Cursor is built on the Visual Studio code, the Open Source editor with Microsoft, and cursor users generate code using AI models such as Google Gemini, Deepseek and Anthropic’s Claude Sonnet.
Several programmers tell Wired that they are now conducting an anthropic coding assistant, Claude Code, next to the cursor (or instead of him). He offers the Claude code since May Different debugging options. It can analyze errors messages, perform solving problems step by step, suggest specific changes and start unit tests in the code.
All this may be beg Is Code written by AI compared to the code written by mistaken people? Earlier this week replication tool for generating AI code Apparently he went dishonest And he introduced changes to the user’s code, despite the fact that the project was in the “freezing of the code” or stopping. It ended with removing the entire user database. Founder and General Director of the Replit said on x that the incident was “unacceptable and should never be possible.” And yet it was so. This is an extreme case, but even tiny errors can get havoc for programmers.
Cooks Zepephere did not have a clear answer to the question whether AI code requires more debugging AI code. Kaplan claims that “it is orthogonal that people often encode the atmosphere.” He says that even if the whole code is written by man, it is still very likely that there will be errors.
Product engineer, Rohan Varma, estimates that among professional software teams, from 30 to 40 percent of the code is generated by AI. This is in line with the estimates made available by other companies; For example, Google said that About 30 percent of the company’s code It is now suggested by artificial intelligence and reviewed by human programmers. Most organizations still make human engineers responsible for checking the code before its implementation. In particular one last one Randomized control test With 16 experienced coders, it suggested that 19 percent took them longer To perform tasks than when they were not allowed to apply AI tools.
Bugbot is to pay this. “The bosses of artificial intelligence at our larger customers are looking for the next step with the cursor,” says Varma. “The first step was:” Let’s raise the speed of our teams, lead everyone faster. ” Now that they move faster, it is: “How do we make sure that we do not introduce up-to-date problems, do not break things?” He also emphasized that Bugbot is designed to detect specific types of logical errors, security problems and other edge cases.
One incident that approved the Bugbot for Anysphere team: A few months ago (human) coders in Anysphere realized that for several hours they did not receive any Bugbot comments about the code. Bugbot collapsed. AnySphere engineers began examining the problem and found a request for a stroke that was responsible for the failure.
There, in the journals, they saw that Bugbot commented on the request to pull, warning the human engineer that if they made this change, he would break the Bugbot service. The tool has properly anticipated its own death. Ultimately, it was man who broke him.