Moves By SpaceX, xAI & Moonshot AI Tarnished By Study On AI Development Tools

A few decades ago, the idea of AI running labs autonomously, advising governments on decisions, or writing code by itself sounded like pure fiction—something reserved for a Black Mirror episode!

Most technology leaders predicted that only by 2050 would AI be embedded into decision-making systems.

Well, here’s the catch: it’s mid-July 2025, and that prediction is showing up in headlines already!

In the last few days, we saw businesses and institutions make moves that felt like early drafts of an AI-dominated future—some exciting, some uncertain, and some raising eyebrows.

So, let’s break down what happened in the world of AI this week!
 

xAI’s “Grok For Government” Program Wins $200M Pentagon Deal

Elon Musk’s AI startup xAI announced on July 14 that its “Grok for Government” suite of AI products is now available via the U.S. General Services Administration (GSA) for all government agencies.

On the same day, the U.S. Department of Defense awarded xAI a contract worth up to $200 million as part of a broader initiative to bring advanced AI capabilities into national defense applications.

Grok for Government includes capabilities such as advanced reasoning, coding, and agentic planning, positioning xAI as a competitor to OpenAI and Google’s offerings.

The announcement came less than a week after xAI’s Grok chatbot faced criticism for generating antisemitic content and controversial political responses, prompting immediate retraining and a version update to Grok 4.

Despite these concerns, xAI stated that the Grok for Government suite is “designed to accelerate America and solve complex challenges in science, engineering, and national security.”

However, that’s not the only shift in Musk’s AI narrative this week.
 

SpaceX Hiring AI Engineers At $170K, Reversing Musk’s Earlier Stance

SpaceX, owned by Elon Musk, recently posted openings for a new AI software engineering team, offering salaries ranging from $120,000 to $170,000 per year, plus stock and bonuses.

Applicants don’t need aerospace experience but should have experience in AI engineering, full-stack development, or data science, according to the job descriptions.

The move is notable because Musk previously stated that AI had no real use at SpaceX or in space exploration, citing poor performance in rocket design questions.

Yet, with the recent release of Grok 4 and increasing cross-pollination across Musk’s companies, SpaceX appears to be shifting focus toward data-driven, AI-enhanced systems.

Musk also hinted that Tesla shareholders may soon vote on investing in xAI, further aligning his business empire with AI initiatives.

While Musk expands his AI bets in space exploration, researchers elsewhere are using AI to speed up breakthroughs in material sciences.
 

North Carolina State University’s AI Lab Discovers New Materials 10x Faster

Researchers at North Carolina State University unveiled an autonomous AI lab that performs dynamic flow experiments, allowing for continuous data collection at half-second intervals.

Compared to traditional steady-state experiments, this new AI-powered system generates at least 10x more experimental data in the same time period.


The setup enables the lab’s machine learning algorithms to make real-time decisions, significantly accelerating materials discovery for clean energy, electronics, and sustainability.

The lab identified optimal material candidates on the very first try after training, while drastically reducing chemical use and waste.

Project lead Milad Abolhasani explains that the innovation “switches from taking snapshots of reactions to streaming full movies”, enabling faster and more sustainable scientific breakthroughs.

As academic labs push the limits of AI automation, the open-source world is also witnessing breakthroughs of its own.
 

Moonshot AI Releases 1-Trillion Parameter Kimi K2 With Agentic Capabilities

Chinese startup Moonshot AI launched Kimi K2, its most powerful model yet, featuring 1 trillion total parameters and 32 billion active parameters in a Mixture-of-Experts (MoE) design.

Kimi K2 is available in two forms:
 

• Kimi-K2-Base: a foundation model for researchers wanting full control for fine-tuning and custom solutions.

• Kimi-K2-Instruct: a post-trained model optimized for general-purpose tasks and autonomous tool use.

The model excels at reasoning, coding, math, and planning, making it ideal for "agentic" tasks—where the AI takes multi-step actions across applications or APIs.


Open-sourced for developers, Kimi K2 aims to democratize access to advanced LLMs, competing directly with proprietary models like OpenAI’s GPT-4 and Anthropic’s Claude Opus.

Moonshot also introduced a custom optimizer called MuonClip, which prevents training instability and makes Kimi K2 one of the most scalable open-source models to date.

Despite all these advancements, not every AI story this week was a win. 
 

AI Tools Made Open-Source Developers 19% Slower, Says Study

A new study by Model Evaluation and Threat Research (METR) found that experienced open-source developers performed coding tasks 19% slower using AI tools like Claude and Cursor Pro.

Despite developer expectations of a 20–24% productivity boost, the real-world results showed that AI often introduced new friction via prompt engineering, code review, and debugging.


Only 44% of AI-generated code was accepted without changes. The rest required human revision, which added to the overall duration of the task.

Researchers concluded that the benefits of AI coding assistants may not transfer well to complex, long-standing codebases where human familiarity plays a larger role than syntactic suggestions.

They noted, however, that newer AI tools like Claude 3.7 may close this gap in the near future as context-awareness improves. However, for now, human developers may not necessarily be faster or more efficient than AI-powered workflows.

From Pentagon’s AI deals to renewed hiring of AI talent to trillion-parameter models and AI-powered self-driving labs, it’s clear that AI is continuing to evolve rapidly. Yet, it’s not without its contradictions, with the METR study showing AI tools actually slow down developers.

So, can xAI’s Grok and open-source newcomer Kimi K2 change the AI race? Will AI tools slowing developers down just add to the technology’s growing pains?

Let us know in the comments below!