As Meta Platforms races to catch up with Anthropic and OpenAI, it’s asking thousands of engineers to use its in-house AI coding agent, MetaCode, in their day-to-day work in an effort to improve the coding capabilities of its AI models.

Maher Saba, vice president of Meta’s Applied AI Engineering organization, last month asked engineers across the company to write at least one code “diff” a week— a code change that is reviewed and integrated into the models, according to an internal memo reviewed by The Information. When the model gets something wrong, they’re encouraged to fix it and submit those corrections.

Saba said engineers’ use and feedback on MetaCode has already helped boost the coding capabilities of the latest version of Meta's AI model, Muse Spark 1.1, and those same corrections will be used to post-train an upcoming model dubbed “Watermelon” internally. 

Meta isn’t just hoping to close the gap between its AI models and those of its rivals. The company is also trying to reduce its reliance on pricey external coding tools, such as Anthropic’s Claude Code and OpenAI’s Codex, after spending billions of dollars on those products this year. Anthropic’s Claude Code has become one of the most widely adopted coding assistants among enterprises, and Meta is among a growing list of companies looking for ways to control the rising cost of AI tools. 

By asking engineers to make at least one coding change a week, the effort goes beyond the kind of product “dogfooding” common across the tech industry, where companies including Google, Microsoft and OpenAI encourage employees to use their own AI products. Meta is also gamifying the push, awarding engineers different-colored badges on their internal profile pages based on the number of coding changes they contribute.

The effort appears to be gaining traction inside Meta. MetaCode now has roughly 7,000 weekly active users and employee corrections have led to more than 800 fixes to the system, Saba said in the memo. Those improvements have also helped boost MetaCode’s score on the DeepSWE software engineering benchmark, according to the memo.

A Meta spokesperson referred The Information to a statement the company has previously given: “It’s well-known that this is a priority, and we’re focused on using AI to help employees with their day-to-day work.”

Meta’s internal push also raises the question of whether the company plans to eventually release MetaCode publicly, as CEO Mark Zuckerberg has hinted. On Meta’s second-quarter earnings call last week, he said the company has “more coding and productivity tools on our roadmap as well.”

The popularity of coding products from Anthropic and OpenAI has made them difficult competitors to dislodge. Open-source coding models have also become increasingly capable, adding pressure on Meta to build its own alternative.

Here’s what else is going on…

Earnings Report

SpaceX’s second-quarter earnings showed how the company is spending heavily to build up its AI business, even as the company reported $7.8 billion in revenue, up 92% from the previous year.

Advanced Micro Devices’ revenue in the second quarter rose 50% from a year ago to $11.5 billion and net profit grew even faster, the company said Tuesday.

Policy Watch

The Trump Administration met with tech companies on Tuesday to share details of a new voluntary framework that will lay out how AI developers should share their models with the government for testing before public release. Axios reported that the White House does not plan to publicly release the framework.

Overheard

Major personal computer makers including HP, Asus and Acer have started using small amounts of chips from Chinese memory maker ChangXin Memory Technologies, Nikkei Asia reported.

OpenAI gave a full-throated response to Apple’s lawsuit filed last month alleging theft of trade secrets. In a blog post on Tuesday, OpenAI called the lawsuit “careless, aggressive and oddly personal.”

Deals and Debuts

See The Information’s Generative AI Database for an exclusive list of private companies and their investors.

Google has assembled a financing program of roughly $200 billion to support Anthropic, with more than $150 billion of it tied to Google's own TPU AI chips and involving firms including Broadcom, Blackstone and Apollo, the Financial Times reported. 

Anthropic struck a six-year, roughly $10 billion deal for computing capacity in Norway, buying it from Volta Infra, an Nvidia-backed AI cloud startup only months old, and bitcoin miner Bitdeer, Bloomberg reported. 

Obsidian Security, which makes cybersecurity software that protects non-human accounts and AI agents businesses run, raised $85 million in a Series D funding round led by Crescent Cove Advisors. 

Decade, a service that pairs each client with a human financial adviser and a proprietary AI model that watches their spending and investments, raised $85 million in a seed funding round from investors including Greenoaks, Benchmark and Diffusion. 

Oligo Security, a cybersecurity company whose software watches applications and cloud systems as they run so it can block hacking attempts in real time, raised $60 million, bringing its total funding to $140 million. 

Convex, a company that provides the behind-the-scenes data and logic layer that developers and AI coding tools use together to build software applications, raised $57 million in a Series B funding round led by Insight Partners. 

QuantHealth, a company that uses AI to simulate clinical trials so drug and biotech companies can test decisions before they enroll patients or spend money, raised $45 million in a Series B funding round led by Qumra Capital. 

Maximum, a startup developing AI software for bank operations, raised $30 million in a seed funding round led by CRV. 

Delightree, a company that sells AI-based software franchises and other multi-location businesses use to run daily operations, raised $25 million from investors including Innovius, Accel, Timber Grove Ventures and Emergent.

Buzz Solutions, a company whose AI software turns inspection photos of power lines, substations and solar farms into useful data so electric utilities can spot problems and plan maintenance, raised $20 million in a Series A funding round led by S3 Ventures. 

New From Our Reporters

Exclusive

ByteDance’s Founder Rules Out Distillation on AI Models

By Juro Osawa and Qianer Liu

China’s New AI Gold Rush: World Models

By Juro Osawa

Thank you for reading the AI Agenda Newsletter! I’d love your feedback, ideas and tips: stephanie@theinformation.com.

If you think someone else might enjoy this newsletter, please pass it forward or they can sign up here.