But the competitive panorama for the coding platforms assisted by artificial intelligence is crowded. Windsurf Startups, Reply and Poolsides also sell generation of code to developers. CLINE is a popular Open Source alternative. Github’s co -pilot, which has been developed in collaboration with Openi, is described as a “couples programmer” which automatically completes the code and offers debug assistance.
Most of these code editors are based on a combination of models to the main technological companies, including Openai, Google and Anthropic. For example, the cursor is based on Visual Studio Code, an Open Source editor of Microsoft and the users of the cursor are generating code by exploiting models for Google Gemini, Deepseek and Claude Sonnet of Anthropic.
Several developers say to Wired who now perform the anthropic coding assistant, Claude Code, together with the cursor (or instead of it). From May, Claude Code has offered Various Debug options. It can analyze error messages, make the resolution of step-by-step problems, suggest specific changes and perform unitary tests in the code.
All this could submit the application: as a buggy AND Code written by artificial intelligence compared to the code written by fallible humans? At the beginning of this week, the tool of generation of code to reply According to reports, it has become a rogue and made changes to a user’s code despite the project is in a “freezing of the code” or pause. He ended up eliminating the entire user database. The founder and CEO of Replit said on x that the accident was “unacceptable and should never be possible”. Yet it was. This is an extreme case, but also small bugs can cause chaos for programmers.
Anysphere had no clear answer to the question if the code Ai requires a greater debugity of the AI code. Kaplan claims that it is “orthogonal for the fact that people are vibrating a lot.” Although all the code is written by a human being, it is still very likely that there are bugs, he says.
The engineer of Anysphere Rohan Varma product estimates that in professional software teams, up to 30-40 percent of the code, is generated by the AI. This is in line with estimates shared by other companies; Google, for example, said it About 30 percent of the company code It is now suggested by AI and revised by human developers. Most organizations are still making human engineers responsible for controlling the code before it is distributed. In particular, a recent one Randomized control study With 16 expert programmers they suggested that it took 19 percent longer To complete the activities compared to when they were not authorized to use artificial intelligence tools.
Bugbot aims to enhance it. “The heads of artificial intelligence in our older customers are looking for the next step with the cursor,” says Varma. “The first step was:” We increase the speed of our teams, we all move faster. Now that they are moving faster, it is: “How can we assure us that we are not introducing new problems, we are not breaking things?” “He also stressed that Bugbot is designed to identify specific types of bugs: logical bugs, security problems and other cases on board.
An accident that validated Bugbot for the Anysphere team: a couple of months ago, the (Human) programmers of Antsphere have realized that they had not received comments from Bugbot on their code for a few hours. Bugbot had gone down. Anysphere engineers started studying the problem and found the request for pull that was responsible for interruption.
There in the registers, they saw that Bugbot had commented on Pull’s request, warning a human engineer who if they had made this change would have broken the bugbot service. The instrument had foreseen its disappearance correctly. In the end, it was a broken human being.
Update: 24/02/2025, 15:45 EDT: Wired corrected the number of Anysphere employees.
Be First to Comment