In just two days, Google and OpenAI unveiled what might be the next evolution of artificial intelligence: the AI agent, signaling a shift away from traditional text interfaces. At its annual I/O event, Google introduced Project Astra, a cutting-edge research initiative where DeepMind engineers are developing a universal AI agent designed to perform a variety of tasks seamlessly. Astra operates through a smartphone, utilizing its AI capabilities to analyze camera feed data and respond to questions about objects in the user’s vicinity with natural-sounding audio.
Hot on Google’s heels, OpenAI showcased a revamped version of ChatGPT just a day earlier, transforming it from a simple text-based interface into a highly interactive assistant tool. Users can now engage with ChatGPT using their smartphones or desktops, asking questions in a conversational voice while the AI processes responses in milliseconds. Remarkably, users can direct ChatGPT to analyze objects, such as text or drawings, and have dynamic interactions, demonstrated in real-time sessions with cleverly drawn math problems. Unlike Google’s exploratory project, OpenAI plans to roll out consumer-ready features imminently.
The driving force behind these advancements is the progress in foundation multimodal models. Google’s AI agent is powered by its next-generation Gemini 1.5 Pro, leveraging innovative techniques developed by DeepMind across multiple modalities, including video and images, to enhance Astra's comprehension of the environment. Meanwhile, the new ChatGPT utilizes the GPT-4o model, which boasts faster processing speeds and enhanced reasoning capabilities over its predecessor, GPT-4. These technological breakthroughs enable companies to adopt a more consumer-focused market strategy, moving away from the lofty, long-term objective of achieving artificial general intelligence (AGI).
So, why the pivot from AGI research to enhancing consumer experiences? According to Bradley Shimmin, chief analyst for AI and data analytics at Omdia, it mirrors the dynamics of the browser wars in the late ’90s. “By integrating generative AI directly into smartphones and applications, from spreadsheets to web search functionalities, leading AI developers like Google can access a vital resource: customer data,” Shimmin explains. This data, instead of primarily generating ad revenue, will be pivotal for refining AI models, allowing firms to develop more effective solutions based on real user interactions.
Eden Zoller, chief analyst of Applied AI at Omdia, added that OpenAI's launch was a model of effective public relations. However, it also raised concerns about accountability. “While GPT-4o's multimodal abilities—enhancing text, audio, and visual interaction—outline exciting pathways for innovation, they also introduce new safety and data privacy challenges. Framing GPT-4o as a friendly companion fosters user trust but risks creating dependency on the AI, particularly if it occasionally disseminates inaccurate information,” cautioned Zoller.
Shimmin noted that emotion-driven interactions in Google's Astra demos reflect a broader industry trend: AI evolving from mere assistance to companionship. “Interestingly, research into prompt engineering has shown that emotionally nuanced queries can significantly improve model performance, a phenomenon echoed by OpenAI’s recent developments with GPT-4o,” he pointed out.
While initial reactions suggested the unveiling of a groundbreaking GPT-5 model, experts clarified that GPT-4o is more of an incremental update with key improvements. Alexander Harrowell, principal analyst for advanced computing at Omdia, remarked, “The focus on enhancing multimodal capabilities aligns with current trends, although the announcement revealed that what was previously thought to be a native multimodal model was, in fact, a pipeline of different models.”
The concept of the AI agent isn't novel, with industry leaders like Yann LeCun, Meta's chief AI scientist, envisioning a future where agents facilitate every digital interaction. OpenAI’s endeavors in this area have been evolving, especially since CEO Sam Altman expressed dissatisfaction with GPT-4’s abilities and the need for a more robust AI system.
Both Google and OpenAI aim to engage consumers directly, demonstrating their AI agents tackling everyday tasks. OpenAI recently announced that new features would be accessible to all users, removing paywalls that previously restricted access. Shimmin believes this shift will significantly benefit OpenAI. “Expanding accessibility not only enhances user engagement and data collection but also positions OpenAI closer to being a consumer-oriented platform akin to a search engine, rather than solely a generative AI provider,” he noted.
The future of AI agents is shaping up to be transformative, bridging the gap between advanced technology and everyday usability, and marking the beginning of a new era in human-technology interaction.
Ranking
