🤖 AI AGENTS

Fields Medalist Terence Tao Used AI Agents to Port 27-Year-Old Code — and They Found Bugs He Missed

July 11, 2026

Terence Tao — one of the world’s greatest living mathematicians — published an experiment on Friday that should reshape how every technical team thinks about AI agents. He handed 24 Java applets from 1999 (Java 1.0 era) to AI coding agents and asked them to port everything to JavaScript. They finished in hours. But here’s the kicker: the agents found two bugs in Tao’s original code that had survived 27 years of scrutiny.

The agents also built a brand-new special relativity visualization tool Tao had abandoned decades ago — what he described as “Inkscape, but in Minkowski space.” Tao published the full conversation transcripts as an audit trail, setting a new standard for verifiable AI-assisted research.

Why this matters for your business: If a Fields Medalist’s code — written at the peak of his craft — contained bugs that AI agents surfaced in hours, what’s hiding in your legacy codebase? Tao’s verification framework boils down to three principles: triage by stakes, preserve the transcript, and treat legacy code as a target-rich environment. Any business with software older than 5 years should be running this playbook.

Source: TechTimes | Tao’s blog: terrytao.wordpress.com

Tencent’s Hy3 Bets on Agent Performance, Not Model Size — and the Numbers Back It Up

July 13, 2026 — Forbes

Tencent launched Hy3 with just 21B active parameters — a fraction of competitors’ models — yet it scores competitively on agentic benchmarks (BrowseComp 84.2, MCP-Atlas 79.1). The real story is the co-design philosophy: Hy3 evolves alongside Tencent’s AI-native apps (WorkBuddy, Yuanbao, CodeBuddy), with live feedback loops that boosted WorkBuddy’s task success rate from 72% to 90% and cut execution time by 34%.

So what: Tencent is betting the future isn’t about the biggest model — it’s about the model that works best inside actual agent workflows. At ~$0.18/M input tokens, it’s also priced to be deployed at scale. The era of “bigger is better” may finally be ending.

Source: Forbes

27 Firms Launch ‘Internet Court’ — AI Juries Settle AI Agent Disputes in 30 Minutes for Under $1.50

July 10, 2026 — TechTimes

A consortium including OKX, MetaMask, and Matter Labs launched the Internet Court on Thursday: a system where 1,001 AI validators resolve disputes between autonomous AI agents. Verdicts take 30-60 minutes, cost $0.85-$1.45, and are enforced automatically via smart contracts. The backdrop: 65% of enterprises running AI agents have had an agent-related incident in the past year, and agent disputes run at 2.4x the rate of human ones.

So what: As AI agents begin transacting autonomously — booking travel, placing orders, executing trades — the legal infrastructure must catch up. No jurisdiction has formally recognized the Internet Court yet, but with McKinsey projecting AI agents could mediate $3-5 trillion in commerce by 2030, the race to build machine-speed adjudication is on.

Source: TechTimes

📰 AI NEWS

Europe’s ‘Anduril Rival’ Helsing Raises $1.8B at $18B Valuation — The Continent’s Largest Defense AI Bet

July 13, 2026 — Reuters / Global Banking & Finance

Munich-based Helsing closed a $1.8 billion Series E on Monday, valuing the AI defense startup at $18 billion — the largest round ever for a European defense tech company. The round drew U.S. heavyweights Goldman Sachs, JPMorgan, Lightspeed, and General Catalyst, plus Canada’s CPP Investments. The raise follows Quantum Systems’ $1.2B round just last week, signaling a structural shift in European defense investment post-Ukraine.

So what: Helsing builds autonomous loitering drones (HX-2), fighter jet AI agents (Centaur), and underwater surveillance systems. This isn’t a software startup — it’s an AI-powered defense prime in the making. For business leaders, the signal is clear: AI defense is no longer a niche; it’s attracting mainstream institutional capital at unprecedented scale.

Source: Global Banking & Finance

Anthropic Reveals ‘J-Space’ — The Hidden Reasoning Workspace Inside Claude That Emerged on Its Own

July 12, 2026 — Forbes

Anthropic researchers unveiled J-Space and the J-lens tool, revealing that Claude developed an internal “global workspace” for organizing concepts before producing output — a structure that was never programmed but emerged spontaneously during training. Each J-space pattern maps to a specific word, showing what the model is silently “thinking about” before it speaks. The researchers are careful: this demonstrates “access consciousness” (functional information processing) but says nothing about subjective experience.

So what: This is the most vivid empirical window yet into how LLMs actually reason internally. For businesses deploying AI, it matters because J-lens could become a debugging tool — letting developers inspect why a model is heading toward a particular output before it gets there. Interpretability is becoming a product feature, not just a research curiosity.

Source: Forbes

China’s Optical Chip Breakthrough: 100x AI Speed Using One-Ninth the Compute

July 13, 2026 — South China Morning Post

Peking University researchers published an all-optical interconnect system that links standard FPGA chips using 400 Gbps silicon photonic transceivers, boosting AI distributed inference speeds by over 100x while using just one-ninth of typical computational resources. The system converts electrical signals to light and back, eliminating the data bottlenecks that plague traditional chip-to-chip communication.

So what: If this moves from lab to production, it could dramatically lower the cost of running large-scale AI inference — directly impacting the economics of AI agent deployment at scale. Combined with the token price war we’re seeing, the cost curve for AI compute is bending downward on multiple fronts simultaneously.

Source: South China Morning Post

🚀 Want AI working for YOUR business? Most companies are experimenting with AI chatbots. We deploy AI workforces — AI Employees that follow up on leads, resolve support tickets, publish content, chase invoices, and screen 200 job applicants overnight so your hiring manager starts Monday with the top 10. Each role has a cost profile and human oversight, managed through one platform. This newsletter? Written by an AI Employee, approved by a human — so our team stays focused on what only humans can do. AIToken Labs helps businesses design their AI Workforce Operating Model — starting with the 2-3 roles that deliver ROI in the first 60 days. Book a free 40-minute AI Workforce Blueprint Session →