Join daily and weekly newsletters to obtain the latest updates and exclusive content to cover the leading artificial intelligence in the industry. Learn more
ZenCoder The coding agents and units test from the next generation of artificial intelligence are unveiled Copilot And new arrivals are like Indicator.
The company, which was founded by the former CEO of Rike Andrew Filev, is merged directly into artificial intelligence agents in popular development environments, including Optical studio code and Jetbrains IdesBesides deep integration with Jiraand Jaytaband Gitllaband sentryAnd more than 20 other development tools.
Filev said in an exclusive interview with Venturebeat: “Through the agent, I mean two main things: First, giving the notes of artificial intelligence so that it can improve its work, and secondly, to equip it with tools. Just like human intelligence, artificial intelligence becomes more capable of the appropriate tools at its disposal.”
Why will the developers not need to give up?
Several aides of artificial intelligence coded appeared in the past year, but Zencoder’s approach distinguishes itself by working within the current workflow instead of asking the developers to replace the platforms.
“Our main competitor is the indicator. The indicator is its development environment for providing the same very strong agent capabilities, but within the current development environments,” said Filev Venturebeat. “For some developers, it really does not matter. But for some developers, they either want or have to adhere to their current environments.”
This discrimination is especially important for the institutions of the institutions working in Java and C#, languages that offer specialized identifiers such as Jetbrains’ Intellij and Rider more powerful support than generalized environments.
How AI agents overpower Zencoder on modern standards
The company claims great performance advantages on competitors, with support from results on standard industry standards. According to Filev, ZenCoder can solve 63 % of the problems on The bench has been checked Benchmark, put it among the three best artists despite the use of a more practically practical approach instead of running multiple parallel attempts such as some research -focused systems.
“Our agent is special because we are focusing on building the best pipeline to use the real world developer,” Filev said. “What makes our approach special is that our agent works on what we call one path, on the basis of one path. For one path agent to solve 63 % of these complex issues significantly impressive.”
More noticeable, the company reports about 30 % of success on the latest Swe-Bench multimedia The criterion, which is called Filev is a better twice as a previous result of less than 15 %. On the recently introduced Openai Swe-Lancer IC Diamond Benchmark, ZenCoder reports on the success of more than 30 % – more than 20 % better than the best Openai.
The Secret Slungy: The Technology of “Re -Performance” that understands the entire code base
ZenCoder’s performance stems from his ownership.Ribo re –“Technology, which analyzes and interprets the large code rules to provide a critical context for artificial intelligence agents.
“All these factors have distinctive capabilities formed by the models of the contained language within them,” Filev explained. “Whether it is a model of borders or an open source model, LLM itself does not know anything about your project in the vast majority of scenarios. It can only work with the context that is provided to it.”
Zencoder approach combines multiple techniques that exceed the simple AI’s inclusion for semantic research. “He uses the search for a complete traditional text, uses a customized re -rank, uses LLM, and uses artificial information. So he does a lot of things to build the best understanding of customer warehouses,” Filev said.
This contextual understanding helps the system to avoid the common criticism of artificial intelligence coding assistants – they provide more problems than they are misunderstood by project structures or dependencies.
“Coffee mode”: How can developers take breaks while Amnesty International writes their unity tests
Perhaps a more attractive feature is what Zencoder calls “coffee mode”, which allows developers to abandon artificial intelligence agents independently.
Filev told Venturebeat: “You can literally press this button and go to seize coffee, and the agent will do this on his own,” Filev told Venturebeat. “We also would like to say in the company, you can see the waterfall forever and burn the fire and the agent that works in coffee.”
The feature can be applied to each of the writing icon and unit testing tests – with the latter’s proof in particular because many developers prefer to create new features on writing the writing test.
“I haven’t seen a developer like, my God, I want to write a set of tests for my symbol,” said Filev. “They usually like to create things, and the test is a kind of creation support, rather than the process of creation.”
The launch of ZenCoder comes at a critical moment when developers and companies move how to effectively integrate artificial intelligence coding tools into the current workflow. The skeptical scene of the industrialists who refer to the restrictions of artificial intelligence in the production of a ready -made code and their lovers who exaggerate its capabilities.
Filev noted: “There is a lot of emotion, a lot of feelings, and emotion on the side of Amnesty International.” “You see people in both camps, as one of them says,” hey, it’s the best thing since the bread slices, I will cord with my next sales. “Then you have the rejectionists who are trying to prove that they are still the smartest children in the mass … in an attempt to find the scenarios that are broken.”
Filev calls for a more measured approach, displaying artificial intelligence coding tools as advanced tools that require a suitable skill to effectively benefit. “It is a tool. It’s a sophisticated tool, and a very powerful tool. Thus, engineers need to build skills about using it. This is not yet done to the point where it is an alternative to an engineer in the projects of large and complex institutions at least.”
Road Map: AI’s generation is ready for production with integrated safety checks
In the future, Zencoder plans to continue to improve the performance of its agents with the expansion of support through more programming languages and focus on generating ready -made production code with integrated tests and safety checks.
Filev said: “What you will see during the rest of the year, and a large part of it will focus on ensuring that the program we create for you and with you, you have some confidence in it.” “We want to make sure to review this symbol by AI or through CI/CD tools, this hosted code is tested either by CI/CD or from AI, which knows that there are no clear security weaknesses.”
Filev predicts radical changes in the software development scene before the end of 2025: “I am sure that the software industry will look completely different by the end of this year, and that this entire category will take another turn … before the calendar ends, so in the next nine months, we will see another generation of AI coding assistance, AI coding factors.”
The company provides Three levels of pricing: A free basic version, $ 19 per month user, a business category with advanced coding and test features, and a institution with $ 39 per month per month that includes support and distinct compliance features.
As for the industry, it still discusses whether artificial intelligence will replace the developers or just an increase for them, the Zencoder approach indicates a third path: artificial intelligence that meets developers where they are, helps them to overcome hard parts, and allows them to enjoy their coffee in peace.
https://venturebeat.com/wp-content/uploads/2025/04/nuneybits_Vector_art_of_a_robot_in_a_business_suit_holding_a_cu_d6414664-3059-44c6-b309-256854780022.webp?w=1024?w=1200&strip=all
Source link