Agentic systems that actually ship
In the wave of 2026 Apple Intelligence features reaching Europe, the Passwords app and Safari stand out for their practical agentic AI implementations. These aren’t flashy demos. They run in the background, interpreting natural language and executing multi-step tasks without constant user intervention. The catch is hardware exclusivity: only devices with Apple Intelligence capabilities, from iPhone 15 Pro models through M1 and later Macs, can access them.
The Passwords app now offers a single button to update all vulnerable credentials. After detecting breaches, weak patterns or stale passwords, the agentic system generates cryptographically random replacements and applies them via supported web APIs. This eliminates manual navigation to each site. Technically it relies on standardized automated password reset endpoints that sites must expose, currently available for Google accounts and major platforms but not universally. The generated passwords stay within the app’s secure storage, reducing exposure during the process.
Safari’s Notify me function takes a different approach to monitoring. You describe in plain language what change to watch for, whether it’s product restock, price drops or new article publications. The agent refreshes the page at user-defined intervals, daily or weekly, without requiring the browser to remain open. It operates as a background process on powered devices, notifying only when the specified condition matches. The limitation on continuous polling makes sense from a battery and server-load perspective, though it reduces real-time utility.
Democratizing extension development
The most developer-relevant addition is Safari’s ability to create extensions from natural language descriptions. No coding required. You explain the desired functionality, perhaps an extension that extracts and rates recipes with specific criteria, and the system generates the complete extension package. Apple allows code review before installation, which is a necessary transparency measure given the security implications of browser extensions.
Under the hood this likely involves an on-device large language model translating the prompt into JavaScript, a manifest file and necessary permissions for the Safari extension format. The generated code runs within Safari’s sandbox, limiting potential damage. For software engineers this represents a glimpse at agentic code generation that could accelerate prototyping, though the output will require scrutiny for efficiency, edge cases and maintainability. The feature addresses Safari’s historical weakness in extension ecosystem compared to Chrome, potentially by letting users build exactly what they need rather than hoping someone else maintains it.
The economic reality check
These tools arrive at a moment when economists are questioning AI’s long-term systemic effects. Branko Milanovic argues that if AI and robotics fully replace human labor, it would mark the end of capitalism as currently structured. His reasoning spans both Marxist and neoclassical perspectives: without workers generating surplus value or earning wages to create demand, the system loses its foundation. The wealth concentrates among owners of the AI infrastructure.
Milanovic points to existing trends like ‘homoplutia,’ where a small elite captures both high labor income and capital returns, representing about 3% of the US population. He connects this to globalization’s effects, where gains from China’s economic rise disproportionately benefited capital owners while stagnating median wages in developed economies. Applied to current AI investments, this suggests accelerating inequality unless mechanisms like universal basic income redistribute the gains. The agentic features in Apple’s ecosystem provide a concrete example of how AI can automate routine tasks, raising questions about which job categories come next.
Infrastructure breakthroughs and geopolitical shifts
Meanwhile, China’s 2025 demonstration of 1 Gbps laser communication from geostationary orbit at 36,705 kilometers using only a 2-watt laser puts Western satellite systems in perspective. The technical approach combined adaptive optics with a 1.8-meter telescope featuring 357 micro-mirrors to correct atmospheric turbulence in real time, then used mode diversity reception. A multi-plane light converter split the signal into eight channels, with an algorithm dynamically selecting and fusing the three most coherent ones. This raised reliable signal acquisition from 72% to 91.1%.
Such advances matter for software systems because they enable higher-bandwidth, lower-latency links for distributed AI training and edge inference across global networks. They outperform Starlink’s typical LEO performance in raw speed from much greater distance while using minimal power. This capability supports applications from high-definition video to quantum-secure communications. For developers building globally distributed systems, it signals that the underlying infrastructure for AI agents is advancing rapidly through international competition.
This technological progress creates openings for regions like Latin America to close the gap with established AI powers in the US and Europe. Yet it coincides with rising tensions between China and the West over critical AI elements, whether compute resources, specialized hardware or data advantages. The cultural undercurrent appears in unexpected places, like the latest Baki anime season depicting Donald Trump and Elon Musk in humiliating scenarios before the fictional power of Yujiro Hanma, reflecting broader unease with concentrated tech and political influence.
What this means for technical leaders
The convergence of these developments demands clear-eyed planning. Agentic AI like Apple’s can eliminate tedious work in security management and customization, freeing capacity for higher-value architecture decisions. At the same time, the economic warnings suggest we should design systems that augment human capabilities rather than pursue total replacement without addressing resulting societal strains.
From an engineering standpoint, focus on building verifiable, auditable AI components, especially when generating code or automating sensitive operations like password management. The Chinese laser communication success demonstrates that breakthroughs often come from clever combination of existing techniques, adaptive optics and modal processing in this case, rather than raw power increases. As competition intensifies, technical teams need to evaluate infrastructure options beyond dominant Western providers and consider how their work fits into a world where AI’s benefits and risks distribute unevenly.
The 2026 landscape isn’t about revolutionary transformation. It’s about concrete tools that change daily workflows alongside deeper questions about economic models and global power dynamics that no single API or laser system will resolve.