Callista AI Weekly (May 19 - 25)

New AI Use Cases

Real-world deployments of AI saw significant momentum this week, as companies across industries rolled out AI-driven solutions to boost efficiency and customer experience:

• Enterprise Operations: BMW is expanding its use of generative AI agents within its Purchasing division. A multi-agent system called AI Conic helps more than 1,800 employees rapidly search and analyse supplier information. By acting as an intelligent internal assistant, the system has already sped up tender processes and will soon proactively monitor supply-chain data and generate reports autonomously, improving productivity and decision-making in procurement workflows.

• Streamlined Retail Experiences: At the National Restaurant Association expo, Samsung and SoundHound demoed an AI-powered drive-thru system that pairs outdoor digital displays with a voice assistant able to take orders conversationally. A live Burger King pilot showed how AI can automate drive-thru ordering, improving speed and consistency for customers. Inside the restaurant, related AI assistants handle phone orders and even coach staff with on-the-job questions, underscoring AI’s growing role in customer service and training.

• SMB E-Commerce Automation: Shopify launched an AI Store Builder tool that lets entrepreneurs create complete online stores using just a few keywords. The generative AI produces page layouts with imagery and text, drastically cutting website-setup time and lowering the technical barrier for small firms to compete online.

• Enterprise Software & Services: Accenture and SAP unveiled a joint offering called ADVANCE to bring “connected intelligence” to mid-market firms. The pre-packaged service embeds AI-driven insights into SAP’s cloud ERP suite, helping companies accelerate finance, supply-chain and workforce processes in as little as six to twelve months. The move shows how consultancies and software vendors are productising AI for faster enterprise adoption.

• Content-Management Agents: Cloud content provider Box announced Box AI Agents for enterprise content. These agents can search, summarise and extract data from a company’s documents by combining the firm’s internal knowledge with leading AI models from OpenAI, Google, Anthropic, xAI and others. Box is also enabling agent-to-agent interactions across different software platforms, pointing to a future where specialised AI agents collaborate across business systems under robust security controls.

From manufacturing to retail and software, these concrete implementations underscore AI’s growing practical impact. They also highlight a trend toward autonomous agents that handle complex, multi-step tasks – a development with major implications for business productivity.

Major Vendor Updates (incl. new models)

• OpenAI + Ive: Project io
  OpenAI has shifted attention from pure software to embodied AI with the surprise acquisition of io, the stealth hardware studio founded by legendary former Apple chief designer Sir Jony Ive. Announced on 21 May, the all-equity deal—valued at about US $6.5 billion—folds io’s 55-person team and Ive’s design collective LoveFrom into a new division charged with creating a family of screen-free, context-aware AI companions. Management framed the move as an essential step toward the agentic era: instead of typing prompts into phones, users would interact with ambient devices that sense their environment, converse naturally and proactively carry out tasks. For business leaders the message is clear: OpenAI now intends to own the hardware layer as well as the cloud stack, mirroring Apple’s iPhone strategy and opening new B2B channels for custom enterprise agents—think voice-first analytics dashboards or factory-floor wearables that work hand-in-hand with GPT-class models. The acquisition also positions OpenAI to compete head-on with Apple, Google and Meta in next-generation human-computer interfaces, signalling that post-screen workflows could arrive on corporate desks within two years.

• Anthropic’s Next Claude: OpenAI’s rival Anthropic unveiled its latest large model, Claude Opus 4. The model excels at complex writing and programming tasks and can employ self-reflection during responses, hinting at self-correcting AI. Early demos showed the system flagging and fixing potential errors autonomously, underscoring Anthropic’s push for reliable, tool-using models.

• Microsoft Build Highlights – Agents Everywhere:

  We've already released an update about the news that Microsoft announced at Build 2025 here. Please find a short summary below.

  • Azure AI Foundry is integrating Elon Musk’s xAI models, starting with the new Grok 3 large language model, bringing more model diversity to Azure.

  • Agent Service reached general availability, letting businesses deploy multi-agent systems with built-in monitoring and enterprise identity management.

  • Copilot Studio allows organisations to create custom Microsoft 365 Copilot agents tuned with proprietary data, while GitHub Copilot is evolving into a background “coding agent” able to run asynchronous code fixes.

  • A new Model Context Protocol aims to standardise cross-agent communication, reflecting Microsoft’s vision for an interoperable “Internet of Agents.”

• Google’s AI Updates: Also for Google, we've already published a specific article about Google I/O 2025 here. In short, Google rolled out AI-powered accessibility tools in Android and Chrome, including automatically generated image descriptions and enhanced OCR for screen readers. The company is testing an AI-mode search experience and integrating its next-gen Gemini model across Gmail and NotebookLM, illustrating Google’s commitment to embedding AI into everyday tools.

• Meta and Others: Meta quietly postponed the launch of its next large model, code-named Behemoth, to later in 2025 after underwhelming performance gains. Stability AI and Arm released Stable Audio Small, a mobile-optimised text-to-audio model, and Salesforce acquired Convergence.ai to bolster its Agentforce platform, signalling competitive moves beyond the usual Big Tech rivals.

• Chinese AI Giants: Alibaba open-sourced Qwen-3, claiming performance parity with top Western models at lower compute cost. Baidu launched ERNIE 4.5 Turbo and ERNIE X1 reasoning models and made its ChatGPT-style service free to the public, showing China’s tech titans are narrowing the gap and competing aggressively on openness and price.

AI Governance Developments

• United States – Regulatory Moratorium Debate: A House bill proposes a 10-year federal moratorium preventing states from enacting their own AI laws, aiming to avoid a patchwork of regulations. Proponents argue national rules will spur innovation; critics call it a giveaway to Big Tech. Businesses are watching closely as the bill advances.

• AI at the White House: President Trump’s administration has subpoenaed major tech firms over potential political bias in AI algorithms while simultaneously advocating AI and coding education starting in kindergarten, reflecting a mixed but high-priority policy approach.

• Europe – Comprehensive AI Rulebook: The EU AI Act is now in force and will apply in phases by 2026. Regulators issued preliminary guidelines on general-purpose models, clarifying transparency and risk-management obligations, signalling that compliance work should begin now.

• Switzerland’s Targeted Approach: The Swiss Federal Council favours sector-specific tweaks and industry-led guidelines rather than sweeping AI laws. Authorities stress that existing regulations already cover many AI uses, providing firms with oversight while preserving Switzerland’s pro-innovation stance.

• Global Initiatives and Industry Self-Governance: Internationally, the UN and OECD continue to promote voluntary AI principles, and the UK will host a second Global AI Summit later this year. Meanwhile, leading AI vendors are funding joint safety research through the Frontier Model Forum to pre-empt stricter regulation.

Breakthrough Research

• AI Solves 50-Year-Old Math Problem: DeepMind’s AlphaEvolve discovered a faster algorithm for matrix multiplication, surpassing Strassen’s landmark 1969 result. The same system improved solutions to dozens of other unsolved problems, demonstrating AI’s potential in R&D across industries.

• Healthcare – Predicting Cancer Outcomes from Faces: Harvard researchers introduced FaceAge, a model that estimates biological age and survival prospects from a simple photo. In cancer patients, the gap between predicted and chronological age correlated with outcomes, suggesting new non-invasive biomarkers for personalised treatment.

• Public Health – Famine Prevention Modelling: A USC-led project combined satellite data and health records to forecast child malnutrition in Kenya six months in advance with nearly 89 percent accuracy, giving NGOs critical time to intervene. The work illustrates AI’s value in integrating diverse data sources for high-stakes predictions.

• Domain-Specific AI Beats General Models: UK startup Vet-AI’s pet-health triage system outperformed general models from Google and OpenAI on veterinary diagnostic tasks. The result highlights that domain-specialised AI can surpass larger, all-purpose models on targeted problems, encouraging businesses to invest in fine-tuned solutions.

Conclusion

In just a week, we’ve seen AI’s transformative reach: enterprises deploying autonomous agents in core operations, tech giants rolling out powerful new models and agent platforms, regulators grappling with oversight, and research breakthroughs rewriting what’s possible. A common thread is the rise of agentic AI – systems that act with increasing autonomy. Whether it’s an agent auto-generating an e-commerce site, triaging a veterinary case or collaborating on a math proof, these developments promise enormous productivity gains and new capabilities for businesses.

Ready to explore how Agentic AI can transform your organization? Visit us at https://www.callista.ch/agentic-ai to discover how we can guide your journey into this exciting new era of AI-powered productivity.

Sources

• Accenture & SAP launch “ADVANCE” AI-driven solutions for mid-market enterprises (Press Release, 20 May 2025)

• BMW Group integrates generative AI agents in Purchasing (Press Release, 19 May 2025)

• Samsung and SoundHound unveil AI drive-thru ordering tech (NRA Show announcement, 16 May 2025)

• Shopify releases AI Store Builder (Reuters, 21 May 2025)

• Sam & Jony introduce io (OpenAI blog, 21 May 2025)

• “OpenAI forges deal with iPhone designer Jony Ive to make AI-enabled devices” (NPR, 22 May 2025)

• “Jony Ive to lead OpenAI’s design work following $6.5 B acquisition” (TechCrunch, 21 May 2025)

• Anthropic debuts Claude Opus 4 (Company announcement, 15 May 2025)

• Microsoft Build 2025: Azure AI Foundry and multi-agent updates (Microsoft Blog, 19 May 2025)

• Google launches AI accessibility features for Android and Chrome (Google Blog, 18 May 2025)

• Salesforce acquires Convergence.ai (Trade media, 16 May 2025)

• Alibaba open-sources Qwen-3 model (TechHQ, 5 May 2025)

• Baidu releases ERNIE 4.5 Turbo and ERNIE X1 models (Reuters, 25 Apr 2025)

• U.S. House advances 10-year ban on state AI laws (DLA Piper summary, 15 May 2025)

• President Trump administration subpoenas tech firms over “woke AI” (AP, 14 May 2025)

• Swiss Federal Council AI strategy update (Federal Council, May 2025)

• AlphaEvolve solves matrix multiplication problem (DeepMind blog, 15 May 2025)

• Harvard’s FaceAge predicts cancer survival from facial photos (Harvard Gazette, 15 May 2025)

• AI model predicts child malnutrition six months early (USC Viterbi News, 15 May 2025)

• Vet-AI triage model outperforms general AI in veterinary diagnostics (BusinessUp North, May 2025)