What automations and use cases do you use Cloud for in your agency?
One of the big things we're using now is a humanization prompt. I created a statistical model of my tone of voice across all my writing on the internet, then use that model to iteratively produce sentence fragments for content. For each fragment I score the perplexity (statistical deviation) from my own writing and select the lowest‑perplexity options, resulting in content that sounds exactly like me. This lets me generate AI content in my own voice, avoiding the typical AI‑generated feel. You can also use AI to generate drafts and then reformat them to match your tone, which works fairly well despite occasional artifacts that appear only about once every few pages. This dramatically cuts down the time I spend on Q&A and content creation.
Beyond that, I'm thinking about more advanced loops to improve existing automations. For example, you could run a Claude code loop that scans your N8N or Claude routines and tries to make them 10 % more time‑ or cost‑efficient, logging results to a shared file. Over time the system can auto‑research, generate hypotheses, test them, and if it reaches the 10 % improvement target it stops. Running this continuously can re‑jig automation components, reduce module count, convert them into code blocks, and eventually offload them to serverless providers like Modal or Trigger, cutting costs by up to 90 % and making the automations run orders of magnitude faster.
Maker School works like this: you join, then go through a pre‑program where I walk you through my methodology and mindset for building a new business, giving you my high‑level view of how I've approached every business I've run. Next, you choose a path. If you lack automation experience, you enter the automation program, which covers an end‑to‑end process including cloud code. I now have a full course on generating media because I see that as the next frontier now that text is saturated. After finishing that (if you still lack automation experience) you move into the main program; if you already have automation experience you can jump straight into the main program. The main program is simply a daily checklist: each day you show up and I tell you what to do—like 'do this, do that.' As part of the checklist I share a video where I actually perform the task, so you learn by reading and watching, then you apply it to your own business. It's not drip‑fed; you can only advance further if you ask me directly and show evidence that you already have a successful business or similar result. I don't let everyone move forward automatically because most people's biggest obstacle is themselves—they get stuck thinking about future content, feel overwhelmed, and lose focus on the daily actions.
Loops have been interesting to play with. The simplest way to get started: before you go to bed, give your AI agent a goal, and have it loop every X minutes as part of achieving that goal — this solves the problem of a model finishing a goal too early, and lets you trade tokens/time for quality since it doesn't matter if it takes 8 hours while you're asleep. Example: I wanted to scaffold out possible structures for a free community, so I set a goal and had it loop every 15 minutes overnight — over about 8 hours that's roughly 32 independent passes, each consulting the master list of ideas so far and adding a few new ones, ending with a massive list of options. A second major use is loops for software optimization: set up a loop where every X minutes an agent spawns and tries to improve some metric you care about (like page load speed) on a live property — it's essentially 'auto-research in a box,' logging what it's tried before and improving a test metric over time. The downside nobody talks about much: loops burn a ton of tokens — creators pushing loops are often fine spending $2-3K/month on AI tokens, so if you don't have that kind of budget, it can be a rough way to work.
Good stuff. My recommendation for you, Curbman, is that we're not actually training the LLMs—that's a common misconception. What we do is applied AI: we take the work from companies like OpenAI, who build and train the models on vast datasets, and we sell it to businesses. We're not doing any training ourselves; we just give the model examples of what we want it to do, which some people call training but it's different from actual model training. Regarding testing the accuracy of the model as stated in the video, it's simple: set up an accuracy test prompt, run it multiple times (e.g., 10, 20, 50, 100, or 1,000 times) on slightly different inputs with the same prompt, score each run, then average the scores. For example, if you get scores like 93, 96, 95, 92, 97, 95, 93, 97, 97, 99, the average is 95.4, which exceeds a 95% threshold, so you can use the model. To be thorough, you can then test with a dumber model and continue down the list to ensure you stay above the threshold; this can be fully automated. I once set up such an automation for my content writing workflow. In terms of expectations, things usually take longer than you think, but once you get initial traction, results accelerate—it's asymmetrical. For me, I wasn't planning to make over $300,000 this month; I thought I'd make around 190k–200k, but I made 1.5 times that because I'd put in daily inputs long enough to be rewarded when it blew up. As for the biggest pain points in our automation services, idea generation was a major issue until we solved it by repurposing YouTube comments and using tools like '1 of 10' to find trending formats, then applying my industry knowledge to make them unique. One of my highest‑performing videos was a clone of a popular video, but I rebranded it to AI and automation—'Watch me do an AI version of this and sell AI.' Editing is also a big pain point; while you can't fully automate it, content repurposing helps. Creators often treat their main YouTube channel as the source and create short‑form offshoots on other platforms. I've tried to address this using creator rewards, though it's hard to measure attribution—we have around 300,000 views but can't track exactly how well it's working. I'll keep you posted.
Here’s how I’d build a fully autonomous reply bot for a cold‑email campaign that sends 10 000+ messages a day. I’d start with a cheap inference platform such as N8N and create an AI agent inside it. The agent would have access to a knowledge base about the company and a library of proven reply templates collected from the past few months.
When a new inbound email arrives, a workflow would trigger that (1) pulls the entire conversation history with that prospect, (2) checks the knowledge base to confirm the company context, and (3) scans the template library to find the closest match for the desired outcome (e.g., booking a meeting). The selected template would be populated with any dynamic details, and the AI would generate the final reply while staying as close to the template as possible to avoid sounding overly AI‑generated.
The idea is to give the AI minimal freedom—just enough to fill in variables—so you retain control over tone and quality. I originally built this in N8N, but once you scale to 100 000 emails per day you can replace the visual builder with lightweight Python scripts running on a serverless platform like Modal, listening to webhook events from your email service.