Subscribe to my newsletter → https://www.sandeepswadia.com/newsletter Most people still use AI like a better search box, but the real shift is AI agents, systems that decide the next action, not just the next word. I explain the difference between prompts and agents using ARR, show what’s happening “under the hood” with four roles, and map how agents adapt through an OODA loop when workflows break. I also cover why agents fail in real life: they amplify vague thinking and bad processes, so you need a GPS check before automating anything. The opportunity isn’t broad intelligence; it’s narrow, specific agents that solve repeated, hated tasks, as output gets cheap and judgment, taste, and standards become more valuable.
ADVERTISEMENT
Training the agent, making sure they don’t fail seems like more work than just going ahead and doing it yourself.
The OODA loop framing is what clicked for me here. A lot of AI discussion treats agents like magic productivity upgrades, but this makes them feel more like delegated operators inside a decision system. They observe, orient, decide, and act, but the quality of that loop still depends on the structure around them. If the workflow is vague, the agent does not fix the ambiguity. It accelerates it. If the standard is unclear, the agent does not create judgment. It performs whatever broken proxy you gave it. That feels like the real management lesson. AI agents make bad systems more visible because they force the organization to say what “good judgment” actually means.
Thank you Harsha Bhogle for such an informative video.
Video Notes Below: 🤖Q: What makes AI agents fundamentally different from traditional AI prompts? A: AI agents are autonomous, recurring, and reviewable tasks that decide the next action rather than just the next word, operating through their own OODA loop (observe, orient, decide, act) to adapt when workflows break and find alternative paths when obvious solutions fail. Agent Architecture and Failure Patterns ⚙Q: What causes most AI agent failures in practice? A: Agent failures stem from vague human instructions rather than bad models, as agents formalize, not fix, bad human thinking and will execute wrong things faster and with more confidence when given sloppy directions, mirroring and amplifying the quality of human thinking. 🔍Q: What is the GPS check and why is it critical before deploying agents? A: The GPS check validates the goal, proof of success, and clear steps before automation, preventing agents from amplifying vague thinking and ensuring well-defined workflows that agents can reliably execute. Strategic Opportunities 🎯Q: Where does the real opportunity lie in AI agent development? A: The opportunity exists in narrow, specific AI agents that solve repeated, hated tasks and understand one workflow, one market, and one user pain better than anyone else, rather than pursuing broad general intelligence. Value Shift in AI Era 💎Q: What skills become more valuable as AI output becomes cheap? A: Judgment, taste, and standards become more valuable as output becomes cheap, with the most valuable person being able to define good work, spot bad work, and know when to trust an agent versus a human. Core Agent Structure 🔧Q: What are the four workers that operate around the language model in an AI agent? A: AI agents consist of a language model at the center surrounded by four workers: analyst, planner, operator, and auditor, enabling the system to adapt to changing circumstances and maintain quality control.
Anither gem dropped 😊 That analogy with the car driver was brilliant
Please create a course expanding this video. Even is it’s a 2 hour one. Would be highly beneficial. Pick real industry example
Humans were created like all other species to survive, not to be productive. AI can increase productivity but that just means it’s optimizing and increasing production. If we focus on automation of impactful repeat work, that’s going to be the winning key. So that whole 80/20 thing? Most agentic thinking right now seems to be focused on the 80% which is just the annoying output stuff (examples: email, triage, ticket process) while that 20% of the more thoughtful impactful repeat analytically derived actions would be most helpful in using agents for. Where does that leave us? Let’s not fear, let’s become more innovative, inventive, creative and really use our mind for even greater impact.
11:11 "I write a newsletter once a week... " yeeeeeaaaaah... 😂 Love the video: very informative. Thank you, sir
You are such a remarkably kind and erudite man. God bless you!!
Your videos are the best videos on AI by far. Thank you so much,
This video is excellent! I understood almost all of it I think and I'm sleep deprived and stoned asf! You're editor deserves a pay kick with the graphics displaying while you're talking and the slow controlled movement's of the videos and things I don't really understanding wtf is going on to do it all. Watched it over a few times but at least I got some of it I think. 🙏💪🙌
First time watching your channel. I love the approach, clarity and excellent analogies. I also think your “calm, positive, monk-like vibe”, is a welcome relief from other creators addressing AI topics. Thanks for your work 🙏
I love learning new things everyday.
I love the way you teach! Thank you ❤
Interesting. I created not an Agent exactly but a GPT App as we call it. It is a customized prompt that is roughly 15 paragraphs in length. What it does is takes anyone’s code and Documents it at first a level to present to a non-technical leadership and then it can do a full technical breakdown for someone new to be able to take it over. Because it is setup and formatted in a specific way all of our documentation is uniform and very easy for anyone to understand. I built it when I realized I had to document 35 ETL scripts and didn’t want to spend 4 weeks doing so. Took about 2 days to get the script right for the ETLs then in order to scale it to the group we did major peer reviews and UATs and still do periodic iterations. However this has saved 100s of hours for our Analysts, Data engineers and Data scientists as it works SQl, to Python to R to SAS.
13:06 Thank you! I love you too!
So Harsha Bhogle is now an AI expert? 🤔
I am so humbled to have added something in my kwonledge of Agentic AI today again, thanks for this.
Damn ❤I truly love you from the bottom of my heart. The way you explain everything is amazing, and I feel so blessed that your channel exists. I’m a student, and after watching your videos, I honestly feel inspired. But the main problem is finding real platforms where we can actually learn effectively. After watching YouTube tutorials, it feels productive, but when it’s time to build something by ourselves, we suddenly can’t do it. In this AI era, nobody really tells us which platforms are genuinely worth following. There’s too much knowledge and too many tutorials, and honestly, it becomes overwhelming and mentally heavy sometimes. Please reply. Love you.
Harsha Bhogle has come a long way from cricket commentary to becoming a AI guru, hats off ... very inspiring