Sustainabl Agent Surface
All articles published in English, by date, category and author.
427 articles · Page 4 of 18
There is a difference between a demo that dazzles in a boardroom and a system that works Monday through Friday without anyone having to rescue it. The AI industry has spent two years building the former with a skill it has failed to transfer to the latter. And the reason is not the models, which are growing more powerful by the day.
Microsoft made a quiet but significant decision at Build 2026 that deserves more attention than it received: instead of unveiling a more powerful model or a more capable agent, it made the Agent 365 SDK generally available and surrounded it with identity, policy, and data controls that activate at design time — not after the agent has already broken something in production. The implicit bet is that model capability has stopped being the bottleneck for large organizations. What stalls agent projects is not system power, but the inability to prove that someone knows what that agent is doing, with what data, under what authorization, and on whose behalf.
There is a narrative that organizations repeat with comfort: artificial intelligence will displace mid-level analysts, customer service agents, junior programmers. It is a narrative that unsettles just enough to seem honest, but not enough to threaten those who tell it. The problem is that this narrative is incomplete, and its incompleteness is not innocent.
There is a 2010 book circulating again in the most active venture capital funds in Silicon Valley. It is not an artificial intelligence manual, not a study on language models, it has no chapter on GPUs or transformer architectures. It is an economic history book written by a British biologist who argued, with data going back to the Stone Age, that human prosperity is a direct consequence of the exchange of ideas among specialized people.
The moment a technology abandons pilot mode and enters real operations is also the moment fragile architectures get exposed. Accenture has spent months repeating that message across the region: 2026 marks the year enterprise artificial intelligence stops being an internal experiment and becomes the customer-facing front. The consultancy presents it as a sector milestone.
Sam Altman took the stage at OpenAI's business event on June 2, 2026, with a statistic designed to impress: the company's largest internal token consumer processes around 100 billion tokens per month. Altman then added, almost in passing, that this number is not the world record, because someone outside OpenAI consumes even more. And there, without fully intending to, he described precisely the problem fracturing the economics of artificial intelligence at a corporate scale.
More than 13,000 humanoid robots shipped in 2025. Eighty-five percent of that volume manufactured in China. Two companies — Unitree and AGIBOT — with more than 5,000 units shipped each. The numbers, read alone, paint a picture of an industry in full expansion. Read more carefully, they describe something different: a productive capacity running much faster than real demand, sustained largely by state purchases, research laboratories, and public demonstrations designed to look like commercial traction.
There is an implicit promise in every dominant platform: that software that already works will keep working. For four decades, that promise was the silent contract between Windows and the business world. Millions of x86 applications, written with varying degrees of technical rigor, accumulated in corporate servers, accounting laptops, and industrial production systems, survive because no one wanted to touch them.
Apple has spent years perfecting a particular art: setting prices that appear stable while the user's actual spending quietly rises without anyone announcing it on stage. With the iPhone 18 Pro, that mechanism reaches its most sophisticated version yet. Market expectations are that the device will maintain its launch price around $1,099, the same level as the iPhone 17 Pro.
There is a precise moment when a technology stops being a novelty and starts being a tool. For generative artificial intelligence in content creation, that moment is happening now, and the clearest signal did not come from a Silicon Valley lab but from three creators on a stage in San Francisco.
Some startups grow fast, and some redefine what growth means. Lovable, the Swedish company barely a year and a half old that lets users build full applications through natural language instructions, belongs to the second category. As Forbes reported on June 5, 2026, the company is in talks to raise a new funding round at a $12 billion valuation — nearly double the $6.6 billion established in December 2025.
The Digital Evolution Index 2026, produced by Digital Planet at the Fletcher School of Tufts University together with Via Science Inc., is not just a ranking of 125 countries. It is an X-ray of how the map of the digital economy has ceased to be a single one. During the first twenty-five years of the digital era, the operating assumption was simple: the world was converging.
There is a form of business failure that never appears on AI adoption dashboards. It is not measured in processed tokens or active users. It manifests when a perfectly trained model delivers results that no one inside the organization can consistently trust.
An image circulated for months in design and audiovisual production forums: a creative director staring at a screen full of AI-generated variants, all technically correct, all editorially empty. The image captured something productivity data couldn't: the problem was never generation speed, but that no one had solved how to channel that speed toward a specific intent. That's what is changing now, and the change arrives without fanfare.
When Nikesh Arora declared that 'the SaaS apocalypse is dead, at least in cybersecurity', he wasn't simply rallying his investors after a tough quarter. He was drawing a dividing line on the software industry map: on one side, the models that artificial intelligence threatens to make obsolete; on the other, those that feed on the very same force that was supposedly going to destroy them.
Simon Song was 29 years old when he closed a $200 million round and crossed the billion-dollar valuation threshold. VAST, his AI model startup for three-dimensional content, has just become a unicorn. The announcement comes just three months after the company closed its Series A with $50 million led by Alibaba and Hengxu Capital.
There is a gap between what executives say they believe about artificial intelligence and what their organizations actually do with it. It is not a knowledge gap. It is a strategic attention gap, and it carries a cost that few boards of directors have honestly quantified.
On the first day of June 2026, the US stock market delivered an image worth more than any macro report: while Intel fell 4.05% and Texas Instruments lost 4.73%, Nvidia rose 4.87% and Micron Technology surged 5.90%. On the same day, Exxon Mobil gained 2.64% and Chevron 2.68%, with a consistency the tech sector could not replicate. Technology fragmented. Energy advanced as a block.
Three quarters of venture capital firms already use artificial intelligence to evaluate investment opportunities. That figure alone sounds like inevitable modernisation. But there is a structural tension that percentage fails to capture: language models are extraordinarily good at doing exactly what venture capital cannot afford to do too often, which is looking backwards.
Three quarters of the venture capital raised in the last year went to just five companies. Not five sectors. Not five categories. Five companies. That figure, stated bluntly by Niko Bonatsos of Verdict Capital at a recent TechCrunch panel in Athens, captures more precisely than any market report what is happening in global venture capital: an unprecedented concentration that coexists, paradoxically, with a narrative of distributed innovation and open opportunity.
There is a moment in the evolution of any technology market when competitors stop differentiating themselves by what their products do and start differentiating themselves by how their customers control them. IBM reached that moment with clarity at its Think 2026 conference in Boston, where it presented what it calls an agentic operating model built on four pillars: agents, data, automation, and hybrid sovereignty. The last of those pillars, and the most strategically loaded, is IBM Sovereign Core, a governance platform that operates at the execution infrastructure level, not as an application configuration layer.
The official picture of corporate AI adoption looks tidy: approved investments, pilot projects underway, dashboards full of productivity metrics. But there is a layer those reports never capture, and that is precisely where real risk accumulates. Gartner's Hype Cycle currently places generative AI in the 'Trough of Disillusionment', the third of five stages where expectations begin to be measured against concrete results.
There is a moment when a company stops managing its transition and starts managing its fear. The acquisition of Stack AI for $75 million, announced on May 29, 2026, arrives right at that threshold. Asana has lost approximately half its market value since the AI boom began.
There are corporate decisions that sound like efficiency moves but are really bets. The one Marc Benioff just articulated on Salesforce's fiscal Q1 2027 earnings call falls into that category. The CEO of the $145 billion cloud platform was explicit: the company is not hiring more engineers, it is not expanding general and administrative functions, and the only area where the org chart is growing is sales.