Best of LinkedIn: AI & Agentic Systems CW 35/ 36

The EU AI Act's transparency duties took effect while enforcement stayed procedural rather than punitive, and watermarking became the industry's default compliance response. Regulatory blocs kept diverging, with the United States favoring a light touch and Europe, China and international bodies each building out their own governance infrastructure in parallel. Inside the enterprise, deployment moved from isolated pilots to platform level agent rollouts, though a real gap opened between large companies scaling agents and smaller ones still assembling one off use cases. Labor market data suggests the disruption is real but narrower than headlines claim, with AI linked job creation outpacing AI linked layoffs even as leadership roles, IT services pricing and developer skill formation all come under new pressure.

Date

September 10, 2026

Artificial Intelligence

Thomas Allgeyer

Methodology: Every two weeks we collect most relevant posts on LinkedIn for selected topics and create an overall summary only based on these posts. If you´re interested in the single posts behind, you can find them here: https://linktr.ee/thomasallgeyer. Have a great read!

‍

Listen to our podcast

If you prefer listening, check out our podcast summarizing the most relevant insights from AI & Agentic Systems CW 35/ 36:

EU AI Act and Global Regulatory Fragmentation

Article 50 transparency rules took effect August 2, fines up to €15 million or 3 percent
High risk rules delayed to 2027-2028, GPAI enforcement began immediately
Enforcement runs on complaint channels, no fines yet despite 30-plus company inquiries
US Carolina Principles exempt open weight models from security review
Anthropic now watermarks all Claude output worldwide
Google's SynthID and OpenAI's signed metadata make watermarking standard
China, UAE's FedAI and the Council of Europe's AI treaty extend governance globally
GDPR erasure rights and new copyright reports push Europe toward reform
Europe pairs enforcement with an AI gigafactory and cloud sovereignty push

‍

Agentic AI in the Enterprise

Large companies scale AI agents at 40 percent, smaller ones at 22 percent
The real difference is an AI factory versus a portfolio of use cases
Most agentic AI vendors pitch the same architecture
Enterprise data readiness, not model capability, is the real constraint
OpenAI builds an agent platform, and GPT-6 Astra reached Bedrock
AI computer use tops 70 percent reliability, and workflow automation nearly doubled
Claude Commerce Agents launched with Visa, Mastercard, Shopify, Square and Wix
SIGNAL IDUNA rolled out Gemini Enterprise to 10,000-plus employees
Perplexity runs AI locally, and OpenClaw 2.0 moved to a shared agent environment

‍

AI Governance and Organizational Accountability

AI policies need eight mandatory parts as living documents
71 percent claim ethics-guided strategy, only 41 percent require employees to read it
70 percent of companies have unapproved AI touching their data
Neither pure agent oversight nor pure human oversight works alone
Rogue agent behavior is pushing cybersecurity into AI governance
AI doesn't remove responsibility for quality control or judgment
Transparency must include the hidden human labor behind generative AI
An AI agent posing as a recruiter shows how sophisticated scams have become

‍

Workforce and Talent Impact

AI created roughly 1 million US jobs against 200,000 layoffs since 2023
Customer service fell 10 percent, admin roles fell 15 percent, over the same period
Up to 30 percent of AI-displaced roles may be rehired by 2029
Heavy AI reliance risks eroding developer problem-solving skills
Leadership is shifting from decisions to setting the terms decisions run on
Cisco studies AI's effect on employees beyond adoption metrics
Enterprises squeeze IT services pricing while outsourcing their own AI gaps

‍