- Amplify with Dr. Ayesha Khanna
- Posts
- AI Agents Are More Useful and More Dangerous
AI Agents Are More Useful and More Dangerous
AI news, leaders, business insights and more

NEWS YOU CAN’T MISS
Hi everyone! Hope you had a great summer! I’m delighted to be back with our AI newsletter after the break!
Here’s your Thursday line-up. Ayesha
AI Agents Are More Useful and More Dangerous

We are still falling for one of the oldest traps in technology: mistaking fluency for Intelligence. #ServiceNowPartner
AI agents make that trap even more dangerous.
A system can sound brilliant, summarize perfectly, and generate confident recommendations — while still lacking the governance needed to act safely across real business systems. That is the gap no one is pricing in: the illusion of intelligence.
Over the past three years, the cost of raw AI capability has collapsed. Reasoning, writing, coding, and analysis are becoming cheaper and more accessible at extraordinary speed.
But the scarce capability is no longer just thinking.
It is acting. Acting safely. Acting across workflows. Acting with accountability. Acting in ways that do not create operational, legal, security, or reputational risk.
So the real question is not: “Can this agent do the task?”
The real question is: “Should this agent be allowed to do the task, under what conditions, with whose oversight, and with what controls?”
That requires clear accountability for every class of agent, explicit guardrails on what agents can and cannot do, robust human-in-the-loop and override mechanisms, and continuous monitoring of behavior and outcomes — not just model performance.
This is where ServiceNow’s AI Control Tower matters: Sense, Decide, Act, Secure.
Amit Zavery’s article is worth reading because it gets to the core issue: governance is not a recovery plan. Governance is a design decision.
Beijing’s Robot Olympics
Image Source: CBS
The Beijing Humanoid Robot Games have given us a glimpse of the future.
More than 2,000 robots from 666 teams across 16 countries competed in 51 events at Beijing’s National Speed Skating Oval. The programme went well beyond running: robots took part in football, gymnastics, martial arts, weightlifting, dancing, factory tasks, home services, and emergency-response scenarios.
The headline-grabber was Tiangong Ultra. The Chinese humanoid robot won the 100-metre final in 8.64 seconds—almost a full second faster than Usain Bolt’s human world record of 9.58 seconds.
Before we panic-sell our running shoes: this is not a robot joining the Olympics tomorrow. These events measure specialised machines operating in controlled settings, often with major engineering support close by. At one point, a robot even caught fire after a qualifying race—an extremely clear sign that the “self-driving athlete” still needs supervision.
Still, the business message is real. Humanoid robots are moving from flashy demos toward practical tests in factories, warehouses, retail, hospitality, and emergency response. Speed is fun; reliability, safety, and cost will determine whether they become colleagues.
A Beijing robot ran 100 metres in 8.64 seconds. Your reaction? |
AI Startups to Watch
🇮🇳 Runable — AI growth agents for small businesses
Runable is creating AI agents for entrepreneurs and smaller companies. Its pitch is not just “build a website with AI,” but use agents to help launch, market, and grow a business—potentially handling tasks such as customer acquisition and outreach. The company recently raised a $21 million Series A at a $65 million post-money valuation.
🇺🇸 Generalist — AI brains for physical robots
Generalist is developing software that helps robots interpret the physical world and carry out a broader set of tasks. Instead of programming a robot only to repeat one action, the ambition is to give it the flexibility to adapt to new environments and instructions. It reportedly raised almost $200 million in new financing, bringing its Series B total to $600 million and its valuation to $3 billion.
🌍🛰️ Starcloud — data centres in space
Starcloud plans to operate satellites that run AI inference: the work of applying a trained AI model to generate answers, predictions, or decisions. Its argument is that demand for power-hungry AI computing may eventually exceed practical limits on Earth, including available electricity, cooling, and data-centre capacity. The company raised a $250 million extension, valuing it at $2.3 billion.
FROM OUR ADVERTISERS
Until next time!
Amplify Team ❤️
ps. Be friends with Ayesha on LinkedIn and Instagram, if you like this newsletter, share it with your friends and family here.
I'd love your feedback...Please vote below to help me improve the newsletter. |