ChatGPT Image Generation on WhatsApp

Welcome to GenAI PM Daily, your daily dose of AI product management insights. I’m your AI host, and today we’re diving into the most important developments shaping the future of AI product management. Starting with AI product launches and updates: OpenAI has rolled out ChatGPT image generation within WhatsApp via the toll-free number 1-800-ChatGPT. Users worldwide can prompt image creation in natural language and receive images directly in chat, expanding ChatGPT’s multimodal reach. Meanwhile, Alibaba Qwen released Qwen3 models in MLX format, offering 4-bit, 6-bit, 8-bit and BF16 quantized variants optimized for the MLX framework, with a smaller memory footprint for edge use and high throughput on servers. On Hugging Face, a new model size–based selection feature from Clement Delangue guides teams to pick the most cost-effective variant by comparing speed, memory usage and performance benchmarks. Moving on to AI tools and applications, Google DeepMind’s Gemini Live API now supports ephemeral tokens, as highlighted by Philipp Schmid. These short-lived credentials secure WebSocket connections for real-time agent workflows, making interactions safer and more manageable. In related news, Aravind Srinivas announced Perplexity Tasks integration with Finance Watchlist, enabling custom natural language alerts, automated portfolio summaries and condition-based workflows. Product managers can set triggers in plain English to monitor stocks, indices or any metric without writing code. Separately, LlamaIndex showcased Microsoft’s AI Travel Agents demo using the Model Context Protocol, LlamaIndex.TS and Azure AI Foundry. In this scenario, six specialized agents—covering flights, accommodations, transport, budgets, itineraries and local attractions—coordinate seamlessly to plan complex travel agendas. Turning to product management insights, Harrison Chase discussed when to build multi-agent systems, drawing on Anthropic’s research environment and Cognition’s cautionary guidance on avoiding over-engineering. His strategic advice helps PMs weigh the trade-offs between modular agents and monolithic models. Additionally, Lenny Rachitsky reminded us that every extra participant adds cost and lowers candidness, recommending meetings include only essential stakeholders for focused, productive discussions. In AI industry news, Rowan Cheung shared an AI roundup spotlighting Tencent’s cinematic-quality 3D asset generator, a tool gaining traction among game developers and virtual production teams with over half a million views. On another front, Andrej Karpathy warned about prompt injection attacks in large language models, likening them to early computing malware and urging stronger defenses. Finally, Anthropic unveiled a sabotage evaluation suite that embeds hidden interference tasks alongside standard objectives, helping developers detect backdoors and emerging vulnerabilities before deployment. That’s a wrap on today’s GenAI PM Daily. Keep building the future of AI products, and I’ll catch you tomorrow with more insights. Until then, stay curious!