This Month in Tech: May 2025

Nobody Codes Here Anymore
AI coding agents like Cursor and Claude Code were rolled out to a team of 40 Ruby on Rails developers. Developers who used the tools were around 20% more productive, especially when increasing code ambition and refactoring. The biggest limitation was remembering to use the agents. While they were great at straightforward tasks, they produced ugly code and required oversight to avoid subtle errors.

How NASA is Using Graph Technology and LLMs to Build a People Knowledge Graph
NASA is using graphs and LLMs to create a People Knowledge Graph. This graph-powered system is built on Memgraph and runs on AWS. It connects employees, projects, and skills to facilitate subject matter expert discovery, project similarity analysis, and real-time organizational insights. NASA ingests data from various sources, extracts skills using LLMs, and uses a RAG-based chatbot interface for querying the graph in natural language.

How Code Reviews are Changing with AI
While AI coding tools get most attention, AI code reviews are having more immediate impact by automating routine checks like coding conventions and test coverage. AI will become better at reviewing existing code before mastering autonomous creation of complex systems. The key shift is moving from “approve first” to “possibly revert later” workflows. Such teams work better after transitioning to a generalist model, where members broadened their skillsets and shared ownership across the entire development process.

Claude 4
Anthropic has released Claude 4, featuring two new models: Claude Opus 4, the world’s best coding model, and Claude Sonnet 4, an upgrade with superior coding and reasoning abilities. These models have extended thinking with tool use, parallel tool execution, and improved memory capabilities when accessing local files.

Gemini Diffusion
Gemini Diffusion is Google’s new LLM. It uses diffusion instead of autoregression for text generation. Initial impressions show that it is exceptionally fast, with Google claiming similar performance to Gemini 2.0 Flash-Lite but five times faster.

Announcing Gemma 3n preview: powerful, efficient, mobile-first AI
Google is releasing Gemma 3n, a new, efficient, and multimodal open model optimized for on-device AI applications.

OpenAI wants to build a subscription for something like an AI OS, with SDKs and APIs and ‘surfaces’
OpenAI wants to develop a subscription operating system for AI with models that ingest every experience users have in their lives. The company hasn’t figured out what exactly that will look like, and it may take a few attempts before it gets it right. CEO Sam Altman says he hopes that whatever the company develops will enable an unbelievable amount of wealth creation. For now, OpenAI will continue building tons of AI infrastructure and keep making models better.

Google’s Next AI Frontier: Autonomous Coding and Visual Discovery Set to Debut at I/O
Google is developing an autonomous software development agent codenamed ‘Codey’ and a Pinterest-like visual discovery feature, both of which are expected to debut at the company’s I/O conference, which starts on May 20.

Apple and Anthropic reportedly partner to build an AI coding platform
Apple and Anthropic have partnered to build a new version of Xcode that will use Anthropic’s Claude Sonnet model to write, edit, and test code. Apple plans to roll out the software internally and hasn’t decided if it will launch it publicly. The latest Claude models have been particularly popular for coding tasks.

Claude can now connect to your world
Anthropic’s Claude AI now offers Integrations to connect with various apps and tools, advanced research capabilities including web search and access to connected applications, and global availability of web search for paid users.

Vibe Check: Claude 4 Opus
Anthropic’s new Claude 4 Opus crushes coding tasks and does something no other AI can - it edits your writing honestly instead of telling you everything is great. While it’s not replacing ChatGPT for daily tasks, Opus one-shotted complex pull requests and built a 3D weather game in 15 minutes. The real standout is its honesty as an editor - it actually tells you when your writing sucks and why, keeping multiple writing principles in mind even with tons of context.

OpenAI introduces Codex, its first full-fledged AI agent for coding
OpenAI’s Codex is an agentic coding tool that allows experienced developers to delegate rote and relatively simple programming tasks to an AI agent to generate production-ready code. It is a unique interface that can be accessed from the sidebar in the ChatGPT web app. Developers can include an ‘AGENTS.md’ file to give the agent custom instructions. Codex is built on a fine-tuned variation of OpenAI’s o3 reasoning model. It is now rolling out to ChatGPT Pro, Enterprise, and Team users - Plus …

Google is about to unleash Gemini Nano’s power for third-party Android apps
Google’s new ML Kit GenAI APIs, likely to be announced at I/O 2025, will enable developers to implement AI features like text summarization into their apps. The APIs will support image input and be available on a wide range of Android devices beyond the Pixel 9 series. They will enable on-device execution of AI foundation models like Gemini Nano to enhance app functionality, keeping user data private by processing data locally.

The Hidden Benefits of Being an Open-source Startup
Open source creates powerful competitive advantages beyond just free distribution. Beyond just differentiation, open source creates multiple competitive advantages: engineers can review code before joining, users provide direct feedback on GitHub, and transparency builds trust with skeptical developers. The approach isn’t without challenges – monetization is tricky, support requests can overwhelm teams, and you might compete with your own free product.

DeepSeek’s R1 leaps over xAI, Meta, and Anthropic to be tied as the world’s #2 AI Lab and the undisputed open-weights leader
DeepSeek R1 0528 has jumped from 60 to 68 in the Artificial Analysis Intelligence Index. This positions the model as higher intelligence than xAI’s Grok 3 mini, Nvidia’s Llama Nemotron Ultra, Meta’s Llama 4 Maverick, and Alibaba’s Qwen 3 253, and equal to Google’s Gemini 2.5 Pro. The model has increased intelligence across the board from previous generations without any changes to its architecture. The gap between open and closed models is now smaller than ever.

1000x Increase in AI Demand
NVIDIA reports notable growth with a shift from simple AI to more complex reasoning, driving a massive increase in AI demand. Hyperscalers are deploying nearly 72,000 GPUs weekly, with Microsoft alone seeing a fivefold increase in token generation. Increased demand requires more data centers, termed “AI factories,” despite efforts to reduce model sizes.

Mistral Launches Agents API
Mistral’s new Agents API enables persistent, multi-agent workflows with built-in connectors for code execution, web search, RAG, image generation, and MCP support.

OAuth for Agentic AI
Microsoft outlined how next-gen AI agents will require evolved identity and access standards, as it shifts from reactive assistants to proactive collaborators capable of operating autonomously across systems.