Everything

Presented at Green IO London 2026 on Green Agentic AI — embedding sustainability into the design and operation of AI agents, measuring their impact across carbon, energy and water, and building more resource-efficient intelligent systems.

AI gets you to ninety percent in minutes, and that speed is intoxicating. But the last ten percent, the judgment, the edge cases, the taste, is where the value and nearly all the effort live. Teams underestimate it because the first ninety looked so easy. The…

People can now produce polished work in areas they don't understand. So output stops being proof that someone knows the work. The person who shipped it and the person who could catch its mistakes are no longer always the same person. Skill and output have…

The YouTube Silver Creator Award arrived this week. What started as a passion for exploring technology, AI, sustainability, and emerging innovations has gradually grown into a wonderful community — thank you to everyone who has subscribed, watched, shared, or engaged.

AI learns from the average of everything it saw, so it pulls your work toward the middle: competent, safe, unremarkable. But advantage lives at the edges, in the choices most people wouldn't make. Lean on it for everything and your output drifts to the mean,…

AI removed the cost of producing options. So the work moved: from making the options to choosing between them, and that was always the harder part. More paths to weigh can mean slower, worse decisions, not better ones. Generating ideas isn't the bottleneck…

Any capability you get straight from a public model, your competitor gets too, at the same time, for the same price. The edge was never the model. It's the data, the process, and the relationships the model can't copy. If AI handed you the advantage, it can…

Each automation looks obviously worth it on its own. Add enough of them and you get a system that works but that no single person can explain. Every local call was rational; the whole is opaque. The risk isn't any one decision. It's the sum nobody chose on…

A tool's capability changes in months. Your approvals, processes, and habits update in years. So you're almost always running last year's playbook on this year's capability, and the gap isn't a skills problem. It's a speed-of-absorption problem. The edge goes…

Building your core capability on a model you don't own means your edge now moves on someone else's roadmap: their pricing, their changes, their priorities. Renting capability is fine until it becomes the thing you compete on. Then the question is whose…

The first question isn't only whether it works. It's whether you can undo it. Some moves are reversible. Others quietly become one-way doors: a dependency you can't unwind, a skill your team no longer has, a process you already deleted. Speed is cheap on…

The failure mode with AI isn't trusting it too much or too little. It's trusting it evenly when its reliability is uneven, sharp on some tasks, shaky on others. Flat trust on a bumpy surface is how a system that's right most of the time still hands you the…

elegating to a system you can't see into means you also hand over the error signal: the small friction that used to tell you something was going wrong. You didn't just delegate the task. You delegated the ability to notice it's off course. Where have you lost…

When a tool answers almost any question you pose, the value moves to posing the right one. That's the part AI doesn't do for you, and using it more doesn't build the skill. The people who pull ahead won't have the best answers. They'll be the ones asking the…

Most of our time at the Green Software Foundation is spent on calls — standards, open source, education, partnerships. Being in the same room for our offsite was a reminder of how much of the trust behind distributed work gets built outside the meeting room.

AI gives fluent, finished-sounding answers. Fluency has a side effect: it closes the follow-up question that would have surfaced the risk. The more polished the answer, the less anyone pushes on it. The expensive part isn't a wrong answer. It's the question…

As agentic AI evolves, the model is only one part of the system — the harness around it shapes how the agent behaves. If we engineer harnesses for reliability, security and autonomy, we should engineer them for resource use too. That is Sustainable Harness Engineering.

A metric is a stand-in for something you actually care about. The moment it becomes a target, AI can hit it perfectly, and the tighter it hits the number, the further it can drift from the thing the number stood for. The measure stops measuring what it…

AI has made content creation abundant. Attention is becoming the scarce resource. So I built FeedLens — an on-device lens that lets my LinkedIn feed adapt to what I want to focus on right now.

AI made junior people produce work that looks senior. So the hard part of the week moved: from producing the work to telling whether the senior-looking work is actually right. And that judgment sits with the few people who were already your constraint. You…

When a project took a year, the market pushed back before you finished building. That delay was accidental insurance: it caught the wrong call while it was still cheap to change. Now you can build the wrong thing at full quality in a week, and the pushback…

Pilots run on your best people, your cleanest data, and a problem you picked because it was tractable. Every condition that made it work is one you strip away at scale. A pilot doesn't tell you the thing works. It tells you it works under the best conditions…

What does it really cost to run an AI agent at production scale? Not only in API charges, but also in energy, carbon, water, and infrastructure. In this presentation, Navveen Balani introduces Lean Agentic AI, a practical framework for building AI agents that…

As AI moves from individual prompts to autonomous agents, the sustainability equation changes significantly. Speaking at Green IO London 2026 on Green Agentic AI: Optimizing Cost, Energy and Carbon.