Musmment icon for the first time AI’s agent with a 70 % victory rate on Github Copilot and Standard Swe’s Standard Points

IconAi Affice Assistant revealed, unveiled a new “new”An increase agent“Today, technology is designed to address the complexity of large software engineering projects instead of generating the simple code. The company claims that its approach is a great departure from other artificial intelligence coding tools by focusing on developers developers on the large large code that extends to millions of software instructions through multiple warehouses.

The company also announced that it has achieved the highest degree so far The bench has been checkedAn industrial standard for artificial intelligence coding capabilities, by combining anthropology Claude Sony 3.7 With openai Thinking model O1.

“Most of the AI ​​CODING space, which is clearly a hot sector, focuses on what people call” zero to one “or” coding devices ” – starting with anything and producing part of the programs by the end of the session.” “What we targeted instead is the engineering discipline of software of maintaining large and complex systems – databases, network stacks, storage – code bases that have evolved over many years with hundreds of developers who are cooperatively working on them.”

Founded in 2022, the MUPMAMATION code was raised 270 million dollars In total financing, including a 27 million dollar round, it was announced in April 2024 with a post -money evaluation of $ 977 million. Includes the company’s investors Sustter Hill Venturesand Project indexand Innovation efforts (Led by the former CEO of Google Eric Schmidt), Lightspeed Venture PartnersAnd Meritech Capital.

How to treat the MUPMENT COUPASES Millions Million

What is determined An increase agent Regardless, according to the company, its ability to understand the context through the huge code rules. The agent includes a symbolic context window, 200,000, much larger than most competitors.

“The challenge facing any Amnesty International system, including an increase, is that when it works with large systems that contain tens of millions of software instructions – a typical thing for meaningful software applications – you can not simply pass all as a context of large language models today,” Ditzen explained. “We have trained our artificial intelligence models to make advanced samples in the actual time, with the correct sub -group specifically from the code database that allows the agent to do its work effectively.”

This approach contrasts with competitors who do not deal with the large code rules or asks developers to collect the relevant context manually.

Another discrimination is the real -time synchronization of code changes across the difference. “Most of our competitors work with old versions of the code base,” said Ditzen. “If you and I cooperate in the same branch of the code and make a change, you want to be artificial intelligence aware of this change, just as it will be. That is why we have implemented the actual time to see everyone to the symbol.”

The company indicates that its approach led to a 70 % victory rate against Copilot When competing for the business of institutions.

Why the AI’s “Memories” feature helps match your personal coding style

Ampment Agent includes the “Memories” feature that learns from developers’ interactions to better compatibility with individual coding patterns and their preferences over time.

“Part of what we wanted to be able to present with our agents is autonomy in the sense that you can give them tasks, but you can also intervene,” said Ditzen. “Memories are a tool for the model to generalize your intention, to pick up that when I am in this position, I want you to take this path instead of the path you took.”

Contrary to the idea that coding is a purely sports logic without stylistic elements, Ditzan stressed that many developers are deeply concerned with the aesthetic and structural aspects of their symbol.

“There is definitely a sporty side of the symbol, but there is also an art of coding,” he pointed out. “Many of our developers want to stay in the code. Some use our agents to write all the code, but there is a full range of engineers who are interested in what the final symbol seems and they have strong opinions about it.”

Institutions’ dependence on artificial intelligence coding tools slowed due to concerns about the protection of intellectual property and security. In addition, it focused on addressing these issues with a strong safety structure and an integration of the level of institutions.

“If you will give them this autonomy, you want to make sure they will not harm,” said Ditzen. “We were the first to provide different levels of SOC compliance and all the penetration test associated with it to harden our solution.”

The company also created integration with developers tools such as Jaytaband linearand Jiraand an ideaand Google ResearchAnd Recession. Unlike some competitors who implement this integration on the customer’s side, Mupment deals with these communications in the cloud, making it “easily and consistently shared through a larger team,” according to Dietzen.

An increase agent is generally available for Opposite the code Users start today, with early access to the inspection Jetbrains Users. The company maintains a full compatibility with the environmental system of Microsoft, unlike the competitor IndicatorWhich is a fork against the code.

“At a level, customers who choose the index cancel the Microsoft ecosystem. They are not allowed to use all the additional ingredients for the Microsoft Vs Code to reach its environment, while we maintained a 100 % consensus with Vs Code and Microsoft Ecosystem,” explained by Detzen.

The advanced partnership between human engineers and artificial intelligence assistants

Despite the progress of helping artificial intelligence coding, Dietzen believes that human software engineers will remain necessary for the foreseeable future.

He said: “The arguments about whether software engineering is a good discipline of people who move forward outside the brand today.” “The discipline in software engineering is completely different in terms of formulating and developing these large code rules, and a human vision will be needed for years to come.”

However, it envisions a future as AI can take over more active roles in developing software: “The real excitement about where we can reach artificial intelligence in the end is only going to quality than the code base and evaluating how the code of instructions can develop to make it safer, more used, and more powerful.”

Current accusation code currently Webflowand Pigmentand lemonadeand DDNand DatastaxAnd Gofundment Among its customers. Pricing starts from $ 30 per month user for professional use and $ 60 per a monthly user to use institutions, with a free layer as well.

The launch of the Mushament agent comes at a time when companies are increasingly looking to take advantage of artificial intelligence to improve developers’ productivity while maintaining software quality. While the institutions are wrestling with the preservation and development of complex software systems, you may find tools that understand the nuances of the large code bases a trap among technical leaders who seek to increase the influence of their engineering teams.

Ditzan concluded that “I hope we can already invest in human software engineering alongside machines”, because we will be able to make much better programs, much more, but much higher quality. I think we can launch a great deal of value for humanity by removing all these barriers to the wonderful software. “

Daily visions about business use cases with VB daily

If you want to persuade your boss at work, you have covered VB Daily. We give you the internal journalistic precedence over what companies do with obstetric artificial intelligence, from organizational transformations to practical publishing operations, so that you can share visions of the maximum return on investment.

Read with us privacy policy

Thanks for subscribing. Check more VB news bulletins here.

An error occurred.