Meta has deployed a new internal AI assistant in coding with multiple AI models, including rivals like Anthropic's Claude.
The tool, called DevMate, became available to employees in March and is increasingly being used for more complex coding tasks that another internal AI assistant called Metamate often struggles with.
The addition of Claude and other models shows that despite tens of millions of dollars investing in tens of millions of dollars to build their own AI models, they hope that their competitor models perform better.
“Like many companies, we experiment with a variety of models to help with coding efficiency,” a Meta spokesperson told Business Insider. “We will continue to improve and collect feedback as we go.”
Software companies, including giants such as Microsoft and Google, are increasingly using AI to write code. Demand for coding assistants such as Cursor and Replit. Many rely on Claude for its ability to handle complex, multi-step inferences than alternatives.
Demand is growing not only from companies but also from employees. For example, Amazon recently expanded its cursor internally after several employees asked about the use of coding assistants, Bi previously reported. This boom has reportedly led to humanity earnings of over $3 billion per year.
Humanity did not respond to requests for comment from BI.
Employees say DevMate has made them more productive
DevMate doesn't just help you write code. It can also analyze failed tests, determine what went wrong, and automatically send corrections for the abilities that engineers have long considered the future.
One current META employee said DevMate has cut its workload in half.
“Devmate turns a 30-minute task into 15 minutes,” the employee said. “It's better than Metamate, so bringing multiple steps together means fewer mistakes. This is important for more advanced tasks.”
Internally, DevMate is considered an agent assistant. This means that it can handle multi-step tasks and perform actions on its own. This is similar to tools like Buzzy Coding App Cursor and Windsurf, another coding app that OpenAI recently acquired.
“Cordrama sucks,” another employee said. Referring to Meta's own coding model, this is one of the options available within Metamate. “It's a great coding model to the 2024 standards, but it's not as good as the 2025 options.”
Employees said that using metamate for simpler tasks, such as pulling up a particular data set, relying on DevMate for larger and more complex tasks, such as building an entire program that moves and processes large amounts of data. Metamate does not support video or images and lacks agent functionality.
A Meta spokesman confirmed that DevMate is being used for more complex coding tasks.
Meta launched Metamate last year, the Financial Times reported. In addition to coding, employees can use it to conduct research and draft internal and external communications. At the time, Meta executives said the company wanted to create “the world's best enterprise assistant.”
A Meta spokesperson said the company hopes Metamate will be useful to all employees, including engineers.
Meta's AI drawbacks
A former meta-engineer told BI that Meta's Llama model has advanced in areas such as multilingual tasks and hallucinations (the trend that constitutes AI things), but is still behind some of the features needed to write the best code.
“We're not there yet when it comes to the instruction and multi-step reasoning needed for a real coding agent,” the former meta-engineer said.
These drawbacks add to Meta's urgency to a broader AI strategy. Bloomberg reported that CEO Mark Zuckerberg is meeting with top AI researchers at his home in Palo Alto and Lake Tahoe and is recruiting a new “tension” team.
Meta has completed a $14.3 billion transaction to get almost half of Scale AI, a data label and annotation startup. As part of that deal, Scale AI CEO Alexandr Wang is expected to join Meta and lead the new team.
Zuckerberg predicts that AI will write half of Meta's code within a year.
“Our bet is that next year, perhaps half of the development will be made by AI in contrast to people,” he said recently.
But for now, AI is coming from competitors.
Any hints? Please contact Pranav Dixit by email to AT pranavdixit@protonmail.com Or at the signal 1-408-905-9124. Use your personal email address and unprocessed devices. Here's a guide to sharing information safely.
