URGENT: 5 Remote Agent Tips To Ship Faster, Better Software

you get an agent and you get an agent and we all get oh my god I'm like Oprah for agents. I want to talk about remote agents. I want to talk about different agents and then five tips that I'm using having ran a bunch of them to make them more effective. Cuz look, you could see the amount of dollars you're spending in real time while you're just walking around. Like I can see right I have them on right now. This is a daily upload. I've recorded the same video a few times. Didn't really like it. Fell asleep halfway through. So, if I fell asleep as the person making it, I assure you I would have been a nice calm.com replacement for falling asleep. But that's not the content I'm trying to make here. Let's bring the energy. So, first of all, different types, right? In the future, you will have Discord or Slack open and you'll just tag the agent. You'll be like, "Go do the thing. Go make growth happen. Go scrape all our testimonials. Go through Discord and find new insights. Go through all of our emails. Go through our CRM. Close more sales." and it'll go and it'll generate some reports. It would make some ads. Like it just is crazy where this is going. Now, we're coders, but I think knowing how to make these things because if you're him and you're a marketer, you don't know how to make this. So, you're stuck in gum loop and you're stuck in the little node thing and you're trying to memorize all the n workflows, but you should have just made it in Python, bro. So, we have a leg up, at least for now, and I think that we should take advantage of it. So that's on the creative stuff where it's like, "Okay, cool. I want to go and build this website like Peter Levels. I want to have these little assets that just print cash. I think you need some research agents in there. I think you need some customer analysis, market analysis agents in there. Do it agentically, of course, where you're just manually having the thing chained together and it comes back with the output." But that's what I'm getting excited about. Also, whoa, this dude makes hour and a half long notebook podcasts out of marketing textbooks. So, if you're trying to learn more about marketing, then you could go and do that. Kind of wild, but yeah. So, five tips. We'll cover that in a sec. Other things that are going on uply updated a new version of AISDLC. If you're not familiar, this is a bunch of my prompts that I use. And you can go to Vibe with AI's GitHub account. And you'll see, you won't see this is V2 that's coming out, but you will see is this AISLC and it's got a bunch of prompts in it. If you hit T and then type in debug, you see all of them in there, right? So those are debug ones and you just chain them together. If you want to have prompts for new feature dev, then boom, they're right in here. I made them YML approved for the ability to have the little models tab, but it didn't make sense for this use case because it's a CLE. Cool. Back to the task at hand. That's an update. That's cool. Also, checking out this Colt UI. Horrible name, by the way. That's a tough one, but they got some nice stuff in there. Now, on to the tips. So, in AI coding, if you're a newbie, then this might be helpful. If you're advanced, the ones later will be more helpful. So, when it comes to this, pick an ID. If they're really similar, you can hop back and forth. I hop between VS Code and cursor. When a cursor comes out with something saucy, then I go and use that. And then I'm always using augment for the heavy lifting. It's just better. It has a better context engine. But cursor is pretty dope at just tab tab tabing. So boom, I got them side by side and I'm ripping the whole time. Need something to manage your context. You can use memory bank. Sure, you can use context engine from augment. Again, recommend that one. And this is context of the actual codebase, right? This is in context of external documentation. If you want that, go grab context 7 or lm.txt. One of the patterns that I use is right here. So on the lefth hand side, I can see I have reference and then this is a bunch of different files that I exported from a different open source project where I was like, whoa, this is a nice codebase. I should totally yink that. So I y it. I did said yoinking by going into their codebase and using augment, you guessed it, for their context engine. And I'm like, yo, how do they architect this thing? Yo, how do they do their CDN? Cuz I like the way they do it. How do they do their documents with Menlify? How do they do their package setup? TS config, all these things. So, I'm using that for context in a markdown file. That's one use case, right? And I'm extracting it from somewhere else. You also have LMS.txt. So, that's what these three are. And then JSON, this is from this little doc scraper, but you can use contact 7 if you want. That's cool, too. I recommend using one of these patterns, though, because if you just dump it into the cursor one, it's you're not gonna have a good time, right? Pizza, French fry. You're not gonna have a good time. So, you need something that's managing the context. Pick one, run with it. And then you need prompts. Already covered that one. Prompts are the sauce, whether you're using remote agents, what I just made a video about on the main channel, where context is king, prompts are the jester. Now, promp prompts may be like the prince. Get the king and the prince. And you need those prompts, right? And you tailor them to you. So if I have a prompt, if I go to snippets, you can see this augment instructions one. And it uses alt six, which is this like crazy squig. Can you see that? That squiggly thing, a wild looking squiggle. But then lin for linear. And what it does is it basically copies the clipboard and then ingests the kit or passes the clipboard into this set of prompt instructions which looks like go watch the other video after this for all the details on doing this. But oh my gosh, come on. Excuse me. I'm not editing this. This is not seen in an edit. So you're just sitting right there and you're just enjoying this. Look at that. That is a nice thing. What is going on? I'm losing my marbles. Oh my gosh. Okay, let's go into Auggie. Okay, now we're back. So, what the prompt looks like is this where basically I do that and then lin and then you'll see at the bottom this prompt. Oh my gosh, it's blurry. The bottom of the prompt basically has a bunch of context that's jammed into it and it's yo do this stuff and then it's passing in the clipboard. So in this case, you'd go and you'd copy a ticket from wherever and then you do alt 6 lin. Wham, it goes off to the races and it gives you the context that you need for the task. So if you're using the DB, do this script that loads in file tree, a bunch of instructions around what to do. If you're doing analytics, do this. If you're doing customer app, do that. If you're using web, then do this. So it's running the type checks, running the tests, using piests. And then after you finish that, update linear with a report on what you're doing. So, if you're using MCP, I'm going to start sneezing. But if you're not, if you're using the direct integration from augment, then it'll just go and rip. And then you have the PR and it's path. So, that's a pretty cool prompt as an example. Then follow the software development life cycle. I'm mapping all these to remote agents right now. And it's just following this thing. So, you got a bunch of steps and you go through them. what it looks like with agents is what I'm doing now because you always do agentic stuff manually first otherwise you don't jump onto the third step on a staircase unless you're like five you might skin your knee or something you got to start at level one which is you do it manually level two agentically meaning it's an agentic workflow so you're still just chaining stuff together and seeing the outputs and then three is full agent never go full agent first but yeah this is what I'm thinking is like prompts mapped to each one of these and then you'd have some master that calls home and yeah those are all in branches. So next use strict formatting and linting rough and black for Python and biome just don't even use that one just use biome it's better also don't use lms for basic things like linting you can just do a script in your package.json so if I go here I go to package.json JSON. I go like this. You'll see I got some biome stuff in here. Format fix, right? I got a lot of stuff in here. Scripts are great. Use them. Like I got logs in here. I got a test just for this vector thing on GCP. We got a lot of stuff happening. This is a good looking package.json, right? You got setup blah blah blah. Cool. So, use those. And then NN type safety. Duh. I'm not using TRPC yet, but Pideantic TypeScript. If you're not using TypeScript, you are a Neanderthal. You're Are you What do you do for fun? You make cave paintings and then use established technologies. So, don't be selfish with your stack. Be smart with it. And then that's it for this video cuz I don't think you care about these amazing VI with AI shades. Only brought to you by Claude Code. Only brought to you by VI and Claude Code. And Michael, this guy in our community, he really wants these because look, you could see the amount of dollars you're spending in real time while you're just walking around. Like I can see, right, I have them on right now. I can see Claude Co's running and I just had to take out a third mortgage against a house I don't own and kids in the future not go to college cuz Claude No, but this idea is funny. That's it for this video though. See you. If you liked one, if you learned one thing, make sure you like the video. If you learned two things, subscribe. Join VIP.