This is the Elon Musk Podcast, your daily hit of what is really going on at Tesla, SpaceX X AI, and the rest of the Musk universe. I'm your host Will Walden, and I have covered Elon Musk for more than five years, spent a year on the ground at SpaceX, Starbase during early Starship development, and before this I spent my career as a software developer working with billion dollar companies.
I've also built and sold my own businesses and now I make content and help other people grow their companies. Now on this show I used that experience to break down the news, filter out all the noise, and give you clear context. You can actually use a Gentek AI gives an AI system the ability to take small sensible steps towards a goal. Instead of stopping after one single reply, think of it as moving from a one time answer to a finished task with receipts
that you can actually review. Businesses use it to book travel, organize calendars, clean out spreadsheets, and draft polite emails that follow company rules. And you get outcomes rather than lose suggestions with a record of how each decision was actually happening. And how does it actually work in everyday life. Though in this episode, you'll get a plain English definition, a simple loop that you can picture in your head, and a few real world stories you can adapt at home or at work.
And we're going to talk about what it does well, where it needs clear instructions of how to keep control with approvals, spending caps, and time limits. And by the end, you'll know how to start small, test safely, and grow with confidence. And we're going to get into that right after this short
commercial break. Here is the short definition of agentic AI. It is a system that pursues a goal through several steps, using tools along the way, until it succeeds or runs out of allowed attempts. Regular chat gives you one response. Then you ask it again, and again and again and again. You get a bunch of responses. An agent continues on its own within the rules that you set for it.
That shift from answers to outcomes changes how you design tasks, how you approve actions, and how you judge success. Now picture a helpful assistant with a checklist. You say plan a family weekend that fits a budget, includes a museum visit. Keep Sunday afternoon free. And the assistant breaks the goal into steps, chooses the next step, uses the right apps, writes down what it found, and it decides what to do next. In agentic AI, the assistant is
the model. The apps are tools like a calendar, a map, a price look up, or your notes. In the written record is a log you can read later. System repeats this plan, act and check pattern until it reaches the goal or asks you for help. It may fail and the main parts are easy to remember. First, the goal, which states what done looks like in simple terms such as stay under $600, keep walking time short, and
pick refundable options. Second, a planner which turns the goal into a short list of small steps. Third, tools which are safe buttons the agent can push like search hotels, read a spreadsheet, or add an event. 4th. Memory, which stores notes from each step so the next step does not forget what just happened. 5th. A checker which asks if the last
step helped or hurt. 6th. A stop rule which ends the run when the job is done, when a limit is hit, or when the system needs your approval, or when it completely fails. Now let us walk through a travel example without any jargon. You tell the agent plan a 2 day Austin trip for two adults, keep lodging out of $200 per night, include a live music event, and leave time for some nice Texas BBQ.
Now the planner lays out steps like find dates, look up hotels, check refund policies, search music events, draft the schedule, and prepare a summary for you. The agent goes to work. It uses a hotel lookup tool, writes down choices for you, compares prices for your budget, and drops the top pick onto a calendar for you. And after each step, the checker asks if the plan still fits the rules. Now the stop rule ends the loop when the schedule, budget, and refund notes meet that goal.
Then the agent hands you a clear summary and a log of every single action that happened there. And the style of AI works best on chores with clear rules, repeatable steps, and measurable results. Reading bills, copying key numbers into a tracker, and writing a short status note fits well because each step has a right or wrong outcome. Drafting polite replies to common messages fits well because examples teach the tone
and structure. open-ended tasks with fuzzy goals like make something creative for the website isn't a good one. It's not a good prompt for that. It can drift because the success target is not clear. You guide the agent by setting a very narrow goal. List rules in plain English and name the tools it may use. Now you got to think about control and safety. It's all within this loop, but
not at the end. You give the agent only approved tools like a read only calendar or hotel search that cannot buy anything. You add spending caps, time limits, and a maximum number of steps. You require human approval for anything irreversible, such as purpose purchases, deletions, or messages to customers. You keep private information safe by redacting secrets, limiting who the agent can contact, and preventing it from
pasting data into public sites. These basic moves act like seatbelts, locks, and curfews for a very fast helper, and the helper can do these tasks within seconds sometimes. So you have to have rules. Now there are a few simple shapes for how agents works. One helper handles a straight path, like reading a form, filling a tracker, and writing a summary about it. A planner and a doer split the job, which reduces trial and error and keeps the log knee.
A small team uses a supervisor to route tasks to specialists, like a calendar specialist, a data specialist, and a writing specialist. Many teams do well with the planner and do repair because it stays simple, exposes the decision points, and makes reviews really fast. Now you do not need a big platform to start. Think of tools as small safe buttons with clear labels and clear limits. One button reads a file, 1 button looks up a price, 1 button adds a calendar entry.
Each button returns a simple result that the agent can understand, and the log shows which button the agent pressed, which words it sent into the button, and what came back. If something looks odd, you can replay the run and see exactly where it went off track and you can figure out how to fix that. Humans are still needed at this point with AI. Some are really good. I run an agent sometime to get me the latest news on Elon Musk and it does wonders for me.
Now. The process that it takes is very straightforward. I say check these amount of news sources, check 20 news sources, etcetera, etcetera, and then send me the links. It's very simple. My my agent is very simple. Send me the links, give me a like a three bullet point rundown of what the article is about. See if it's worth my time. I'm going to redevelop this agent into something a little bit more robust. So it's easier for me to do this
podcast. I've been doing this podcast for a very, very long time, five years or so thousand episodes plus. So if you are a fan of the show, thank you, and if you aren't, thank you for stopping by and listening. And also, since you are a new fan to the show, please take a second and hit the follow button. That'll be really helpful. Or pick up some merch at starshipshirts.com.
That helps out tremendously. So agents are really cool because you can measure an agent the same way you measure a person who handles a task. Then it finished the task, yes or no. How long did it take? How much did it cost to run it? How often did it ask for help? Now those numbers give you a clear picture of whether the system saves time or actually creates more work for you.
When I first started developing my agent, I was, it was taking me more time than it than it was before I had it just to get it right. But by fine tuning it, I have a clear path to, you know, the best results for the day. I can't just Google things. You get a bunch of just junk in there and you can't just chat TBT things because it doesn't give you the right answers. So you have to make something.
I had to make something custom. I had to make a custom agent to do the things that I wanted to do. And it has a very simple clear task. And then it doesn't fail anymore because it knows exactly what to do, has like 4 things it needs to do. Search the web on certain sites. Send me back the headline. Send me back three bullet points about the article because it reads the article. Send back three bullet points of the article. Send me a link to the news piece
so I can check it out myself. And that's it. That's all it does. And then I read everything I can and come up with stuff on my own. So my agent is very simple. You can make something absolutely complex. Crazy, you know? How often does it ask for help? That's the important one, right? So those numbers give you a clear picture of whether the system saves time or creates a huge headache for you. And if it creates the headache, either work on it more or find
another process. I had to go through 5 or 6 different processes in order for my agent to work properly. If you have the time, great. If not, there are systems out there that will make you an agent. Or you can do like a drag and drop agent for this kind of stuff. And you have to treat each run like a transaction with a result you can accept or reject because each one of these costs compute time or tokens. If you do it on something, you know where somebody else hosts
the LLM. And when you reject a result, add a quick note about what went wrong and then turn that note into a rule the agent can follow the next time. And let's talk about like a story from an office because that might put it into perspective. So customer support, we all hate it, right? We all hate going through customer support. If something goes wrong, we want to answer now. But if you're on the other end where you're a customer support agent, inbox assistance are important.
And if you have a goal, the goal would say, read each new message, match it to one of five common issues. Suggest a reply that follows the playbook, which is, you know, you have a playbook that you feed it and hand anything unusual to a person. So the agent opens an e-mail, finds key details, checks the playbook, drafts a reply and puts it into a queue for a person to review.
And if the person reviews it and it needs a little bit of help, type in there, you know, sorry, sorry you're having this issue. My name is Will. I'm here to help you. And that person confirms that the draft uses the right issue type and the right tone. So it's basically a huge database of things like the playbook. Now, a person gives a quick approval on the first week of use, then reduces approvals to only high stakes messages once the numbers look steady. So you're training this model
while it's doing its job. Now, another one we can think about is where you mix tools and approvals. So if you're a finance helper and it could check small purchase requests, the goal says ensure the request sits within the budget, confirm the vendor is on the approved list and prepare a draft purchase order. The agent reads the request, looks up the budget balance, matches the vendor name, and
fills a form. The stop rule blocks any e-mail to a vendor until a human clicks approve, and the log acts like a receipt for an audit, which keeps everyone comfortable with the tool that touches their money. You don't mess with people's money, man. So if you want to start this, it's straightforward and there are tools, you have to look them up. I don't want to suggest anything because I'm not very, I'm not very educated in what tools are out there other than the ones
that I build myself. I'm a coder, so I build these things myself. But to start, you can pick a tiny job that annoys you somebody like pulling a number from a document and dropping it into a tracker or drafting a follow up e-mail after a meeting. Google does this a lot. You know the all the e-mail apps do this now, but a follow up e-mail after a meeting? You could have your agent do that for you. Write a goal in one sentence and list 3 rules the agent must
follow. It's super simple, a goal. You could tell the the AI agent every time I get an e-mail, send it to my phone or what you know, like give me a give me an alert on my phone, something like that. Send an alert to my phone number and text you, you know, text you a summary of the e-mail. And if it sees seems like a big deal, then sure, answer it. But if not, just leave it alone.
And I know there are e-mail apps that do that by themselves, but you know, we're just making a, a silly model right now. Or if certain person emails me, send me a text. How about that? That would be great. That's an actual like a list of a actual thing that could be helpful. If my boss emails me, send me a text because sometimes your e-mail alerts don't go through or sometime it's after office hours, but your boss might need to get a hold of you. You don't have your e-mail app
on your phone. Your your work e-mail app on your phone. So you know you do you get a text from your boss. It's 8:00 at night when you're putting your kid to bed. That would be the worst app ever. Don't do that. I, I'll tell you this, do not have your work and your home like and your personal phone as one phone. That's the worst idea ever. Unless of course, you run your own business and then you got to do it. So eventually you can just run it on autopilot and does all the
stuff for you. You can have your agent do so many different things and you can check out different services. Like I said, I don't want to recommend anything because I don't really use them myself personally. I built them, so I build them for my own personal use. So I hope that this has cleared a few things up for you. It's like a helper, you know, if you, if you have an executive assistant, that's kind of what an AI agent, agentic AI is all about.
It helps you do those tasks that you just don't want to do. It turns a chatty program into a helper that completes tasks by planning those steps using tools that are safe and then checking along the way so you can get the results that you want. You just got to give it a clear goal, a few rules and some buttons that it can push. And you keep all the control too. You get readable logs from it. You measure success with finish rates, time to finish and escalations. Like does it send it back to
you? So start small and confirm the value and grow your agentic AI at your own pace. You don't have to make a huge project, make something very small that'll fix a task or fix a problem that you're having right now. Hey, thank you so much for listening today. I really do appreciate your
support. If you could take a second and hit the subscribe or the follow button on whatever podcast platform that you're listening on right now, I greatly appreciate it. It helps out the show tremendously and you'll never miss an episode. And each episode is about 10 minutes or less to get you caught up quickly. And please, if you want to support the show even more, go to patreon.com/stagezero and please take care of yourselves and each other and I'll see you tomorrow.