When AI meets Product: February’25 AI Product Updates

Keeping up to date with new AI models, products, ethics, and trends

One more proof that we are in a new AI era? AI sneaked into Super Bowl ads! Both Google and OpenAI had AI-focused commercials during last month Super Bowl, but with very different styles. OpenAI’s ad was inspirational — showing humanity’s biggest discoveries, from fire, wheels and the moon landing, to ChatGPT. Meanwhile, Google took a more practical approach, showcasing how AI can help small businesses in the U.S. with translation, summarization, and other tasks to help people save time for more important things.

Welcome to the February edition of “When AI Meets Product — AI Product Updates”. This month, on top of AI commercials during the Super Bowl, we saw many interesting AI updates:

• GenAI Model Updates — Big GenAI model providers surprised us with many new model versions and also launched narrow use case implementations on top of them.
• AI Product Updates — Everybody is working on AI agents now. Even companies that before didn’t allow employees to use GenAI are changing their mind, and Duolingo launched an interesting AI video call to help people practice speaking.
• AI Ethics and Legislation Updates — We got some official guidelines for AI Act, but it looks like regulation might cool down in favor of innovation.
• Other resources: what GenAI means for tech teams— Some useful perspectives on what product teams need to think about when building AI products and how AI is changing team structure and strategy.

Let’s get started! 🚀

GenAI Model Updates

Anthropic

• Claude 3.7 Launch — Anthropic introduced its first “hybrid” model, Claude 3.7. This model can switch between instant direct responses and more detailed, step-by-step reasoning that is visible to the user. API users also get control over how long the model spends thinking. Pricing remains the same as previous versions ($3 per million input tokens, $15 per million output tokens), but be aware — thinking tokens are now included, which may increase costs. They also launched a New Developer Tool, Claude Code, a command-line tool designed for agentic coding.
• AI Safety Research — A new paper from Anthropic’s Safeguards Research Team outlines a method to protect AI models from universal jailbreaks.
• AI in the Economy — Another study, based on millions of Claude conversations, explores how AI is being used across different economic tasks: 50% of AI usage is concentrated in software development and writing, 36% of occupations use AI for at least a quarter of their tasks, 57% of use cases show AI augmenting human capabilities while 43% suggest automation.

OpenAI

• No O3, but a Hybrid Approach — OpenAI canceled its planned O3 release in favor of a more unified, hybrid solution (similar to Claude 3.7?).
• Introducing GPT-4.5 — Unlike reasoning models like OpenAI O1, GPT‑4.5 improvement is not based on Chain of Thought reasoning. Instead, its improvements came from scaling up compute and data, along with architecture and optimization innovations.
• ChatGPT Growth — ChatGPT’s subscriber base nearly tripled in 2024, reaching 15.5 million users.

Other Model Providers

• xAI — Launched the Grok-3 family, a set of four text-only large language models, including both reasoning and non-reasoning versions.
• Alibaba — Introduced Qwen2.5-Max, a large-scale MoE (Mixture of Experts) model exploring new levels of intelligence.
• Perplexity — Released Sonar, a new model built on Llama 3.3 70B, designed to enhance search capabilities for Perplexity Pro users.

For an in-depth review of the latest models (Grok-3 and Claude 3.7), check out “A New Generation of AIs — Yes, AI Suddenly Got Better… Again”.

Narrow Applications from the Biggest Model Providers

This month, we’ve seen a growing trend of narrow AI applications — more specialized tools built on top of GenAI models to solve specific problems — coming from the same big GenAI model providers:

• Google — Introduced AI Co-Scientist, an AI-powered research assistant.
• Perplexity & OpenAI — Launched Deep Search / Research, improving AI-driven information retrieval.
• OpenAI — Unveiled a computer-using agent, along with GenAI solutions for government applications.

We also saw some key partnerships between Anthropic & Lyft (developing AI products to assist riders and drivers), Anthropic & Thomson Reuters (creating AI-powered solutions for tax research, productivity, and client service) and Cisco & Mistral (collaborating on AI innovations to enhance customer experience)

This shift toward practical, domain-specific AI shows how major players are focusing on real-world applications and try to solve the user problem when dealing with all-you-can-ask chatbots “yes this is great, but what should I ask / how is this useful?”.

AI Use Cases & Product Updates

Some companies initially resisted AI due to concerns over privacy, confidentiality, hallucinations, and content quality. However, we’re now seeing a shift in perception as businesses recognize AI’s potential to augment human capabilities rather than replace them or decrease quality of their work. A notable example: The New York Times has greenlit AI tools for its product and editorial staff.

At the same time many companies are already deploying agents in production to improve their products:

• OLX —Integrated GenAI and AI agents to enhance search functionality. Check out this video for insights into the product strategy and technical details of these new features.
• Glean — Launched AI agents to help companies achieve better internal productivity and knowledge management.
• GitHub Copilot — Expanded with agents for code completion, chat, and multi-file editing.
• Replit — Released its first software creation agent on iOS and Android, competing with other AI-driven web and app development tools like Bolt, Cursor, Vervel, and Windsurf.

Education continues to be a major area of AI innovation, particularly in personalized learning and feedback. Duolingo recently launched an AI-powered Speaking Video Call, making language practice more interactive. Some interesting product choices include: a sassy and sarcastic AI personality to keep users engaged, and automatic call endings after a few minutes to maintain the “short lesson” format and prevent addiction.

AI Ethics and Legislation Updates

This month brought both new guidelines and shifting global perspectives on AI governance.

• AI Act Guidelines Published — Official guidelines on AI system definitions and prohibited practices were released, marking another step toward AI regulation in the EU. However, as expected, they have also faced criticism.
• AI Summit in Paris — Many interpreted the summit as a signal that multiple countries are looking to ease AI regulations in favor of fostering innovation. This reflects an ongoing debate between maintaining control over AI risks and accelerating technological progress, as well as global political trends.

Other resources — What GenAI Means for Tech Teams

As GenAI becomes more integrated into products, it is already reshaping how engineers implement AI, how teams are structured, and what customers expect from technology. We should expect this changes to continue and even accelerate in the close future.

Implementation: Building AI-Powered Products

• Martin Fowler shares key learnings on implementing GenAI integrations, covering prompts, evals, embeddings, RAG, and fine-tuning.
• Your AI Product Needs Evals — A framework ensuring AI solutions are iterated and improved across three key areas: evaluating quality (e.g., testing), debugging issues (e.g., logging & inspecting data), adjusting behavior (e.g., prompt engineering, fine-tuning, and coding).

Team Structure & Strategy

• Marty Cagan predicts GenAI will impact product teams in several ways: smaller team sizes due to automation, more focus on discovery / less on delivery, expanded scope of what teams can build, cost savings and workforce disruption…
• Karen Stroup emphasizes the importance of board alignment for AI strategy, as well as the importance of AI governance, training teams to work effectively with AI, and balancing innovation with user experience and ROI.
• Brian Balfour outlines 7 Ways Customer Expectations Are Shifting, including the shift from “A Place for Me to Create” to“Do the Work for Me”. This has big implications for product teams strategy and roadmaps.

Wrapping it up

That was it from “When AI Meets Product — AI Product Updates”. 2025 is starting really strong with new AI models, agentic use cases and impact on Product and Tech teams. Stay tuned!