Regulation and Governance

• A new plain-English framework walks businesses through six steps to classify AI systems correctly under the EU AI Act, aiming to avoid both wasted compliance spend and fines
• Framing compliance as a burden misses the point for small and mid-sized businesses: the same governance discipline the EU AI Act requires is what makes agentic AI reliable enough to deploy at scale
• Gartner expects 40 percent of enterprises to demote or decommission autonomous AI agents by 2027, not for poor performance but for governance gaps that only surface after something breaks in production
• A new cheat sheet maps four leading agentic AI governance frameworks, arguing organizations do not need to build oversight from scratch as agents gain the ability to retrieve data, use tools, and take action
• Agentic AI is turning the data layer into the new security boundary: agents that can query, summarize, and act on data need protection built into the database itself, not just the application layer
• Bank of England Deputy Governor Sarah Breeden told the ECB Sintra Forum that existing financial regulatory frameworks were not built to contemplate autonomous agents, and that relying on human-in-the-loop for every action is no longer realistic
• At Geneva's first UN Global Dialogue on AI Governance, the dominant theme was that most governance today assumes a human makes the final call, an assumption agentic AI is starting to break

‍

Sovereignty and the Global AI Race

• OpenAI proposed handing the US government a 5 percent equity stake, worth roughly 42 billion dollars at an 852 billion dollar valuation, with reports suggesting Anthropic, Google, and Meta could be invited to follow the same template
• The US Commerce Department lifted its export controls on Anthropic's Fable models after a brief suspension, restoring international access
• Salesforce committed 1 billion dollars to Switzerland to accelerate agentic AI development, positioning the country as a research and build hub while the Gulf increasingly becomes where agentic systems get scaled
• The UAE is targeting growth from roughly 1,500 AI companies to 10,000 by 2030, backed by Abu Dhabi's MGX, which has reportedly raised nearly 50 billion dollars for AI investment
• Portugal launched AMALIA, the first open-source AI model built for European Portuguese, following Switzerland in treating AI infrastructure as language and public-sector infrastructure
• On OpenRouter, Chinese models crossed 45 percent of total token volume, compared to 15.3 percent for Anthropic and 7.4 percent for OpenAI, even as frontier capability still sits mainly with American labs

‍

Enterprise Adoption and Workforce Economics

• A Deloitte survey of more than 3,200 business and IT leaders across 24 countries found that 61 percent of senior leaders at large US and UK professional services firms abandoned at least one AI project last year, most often because their people lacked the skills to run it
• Only 13 percent of companies successfully scale AI automation, with most attempting to jump straight from pilot to full deployment instead of fixing the underlying process first
• IDC predicts seat-based enterprise software pricing will be effectively obsolete by 2028 as agentic AI reshapes the buying model, while MIT found that roughly 95 percent of enterprise GenAI pilots deliver no measurable P&L impact
• Harvard's Christopher Stanton argues AI will not devalue all SaaS equally: products built on deterministic tasks and internal, undifferentiated data are the most exposed to replacement
• Europe's largest insurer cut up to 1,800 jobs after integrating AI into claims and customer service, a move tied to insured natural catastrophe losses hitting 127 billion dollars last year, 27 percent above the long-term average
• New SAP and Oxford Economics research shows EMEA organizations generating stronger returns on AI investment, with expectations for further gains rising over the next two years
• OpenAI's own research on Codex found that over 80 percent of agent-driven tasks now correspond to work that would take a person more than 30 minutes, and 70 percent to work exceeding an hour

‍

Markets, Funding, and Infrastructure

• OpenAI opened a European headquarters in Dublin, creating 250 new roles over the next two years
• Apple sued OpenAI over alleged theft of trade secrets, a case expected to run for years and carry meaningful risk for OpenAI
• Anthropic is now tracking roughly 74 billion dollars in annualized revenue, adding about 550 million dollars in new ARR every day in June according to Yipit, nearly doubling OpenAI's pace from a base a sixth of its size eighteen months ago
• Europe's AI unicorns are pricing on different bets than the US: Helsing sits at 18 billion dollars on 35.9 times revenue, Legora trades at 56 times, Fluidstack at 41.7 times, and Nscale sits at 15 billion dollars priced like infrastructure at 9.7 times
• A map of the 35 most important AI companies shows the frontier spreading well beyond the Bay Area, with Mistral AI valued around 14 billion dollars in Paris, and London anchoring DeepMind, ElevenLabs, Synthesia, and Stability AI
• Anthropic's Fable 5 topped the new EuroExec Benchmark at 81.4, ahead of GPT-5.5 at 77.0, in a test built around European executive scenarios rather than math or code
• Verizon detailed a Level 4 network autonomy architecture covering 70 million automated changes in 2025, while Nokia and Google Cloud embedded six Gemini-powered agents into Nokia's Assurance Center targeting 50 to 80 percent faster fault resolution

‍

Open, Local, and Edge AI Infrastructure

• A new benchmark found local, private AI models can now match frontier cloud systems on real enterprise workloads at a fraction of the cost, generating a comparable research report for one cent versus 7 dollars
• An offline edge AI system built on NVIDIA Jetson, LoRa radio, and IoT sensors is now delivering decisions in under 1 millisecond for mines, ships, farms, and other sites without reliable internet access
• Open standards groups including LF Edge and the Agentic AI Foundation are pushing for open interfaces and interoperability as agents move from experimentation into production
• Together AI's Vipul Ved Prakash argues agentic AI is entering its modular architecture phase, the same shift that let the PC ecosystem outpace vertically integrated competitors once interfaces stabilized
• Proprietary AI still dominates consumer products on distribution and polish, but open weights are becoming the foundation builders use for fine-tuning, self-hosting, and enterprise customization
• Recent AI scientist breakthroughs, including a two-day antimicrobial-resistance discovery and a machine-written paper that passed first-round review at an ICLR workshop, still depend heavily on human-guided scaffolding behind the scenes

‍

Thanks to Jothi Moorthy, Adam Helbig, Omar Hammami, Christiane Zhao, Niharika Srivastav, FAISAL HOQUE, Bernard Marr, Oliver Patel, Brij kishore Pandey, Nandan Mullakara, Greg Coquillo, Ashish Dhawan, Joanna Rubin, Ali Najafzadeh, Alan Paris, Eric Pilkington, Sandeep S., John Beardsley, Chad Johnson, Manipal Dhariwal, Lou Simon, Adam King, Ting Zou, Agnese Jaunošāne, Dr. Barry Scannell, Soumitra Dutta, Tina Austin and Tony Moroney for contributing insights to this edition. Find the full list of posts and voices on LinkedIn: https://www.linkedin.com/pulse/best-linkedin-cw-29-30-ai-agentic-systems-thomas-allgeyer-tabof?trackingId=x9%2Fg71nbpjqvXTBG2MGs0Q%3D%3D&lipi=urn%3Ali%3Apage%3Ad_flagship3_profile_view_base_recent_activity_content_view%3BMRkLsnt5R46kvOxB%2BHTaAg%3D%3D