Get AI summaries of any video or article — Sign up free
8 AI Agents & Tools I Use to Make $1.6M / Year thumbnail

8 AI Agents & Tools I Use to Make $1.6M / Year

Simon Høiberg·
6 min read

Based on Simon Høiberg's video on YouTube. If you like this content, support the original creators by watching, liking and subscribing to their content.

TL;DR

The business reports near-90% profit margins after shrinking a nine-person team to four by replacing many tasks with AI tools.

Briefing

A seven-figure SAS business owner says the fastest path to near-90% profit margins in 2024 was replacing much of a nine-person team with AI tools—then wiring those tools into four core business functions: operations, customer support, content creation, and product development. Total revenue topped $1M last year, with a target of at least 1.5× growth in 2025, but the more striking shift was headcount: the team shrank to four people while AI took over large portions of day-to-day work.

The workflow starts with AI chat. OpenAI’s ChatGPT remains the most broadly useful “chat with it for a long time” option, while Gemini is singled out for location-and-maps tasks. Alternatives are also in the mix—Claude as a preferred choice for many founders, DeepSeek for strong recent results, and Llama as an open-source option. Cost becomes a deciding factor: ChatGPT’s $200/month plan is described as too expensive for the market, so the business uses OpenAI models directly through a chat UI called TypingMind to get more value.

Automation is handled by n8n, positioned as a more flexible alternative to tools like Zapier or Make because it supports advanced workflows while still letting users drop into code when needed. n8n can also serve as a backend prototyping layer by generating API endpoints via drag-and-drop. The key upgrade is agent-building: connect n8n to OpenAI, add memory and tools, and let an AI agent autonomously decide how to use those tools to complete tasks—reducing the need to handcraft complex conditional automation flows.

Customer support is addressed with 8base, an AI-powered support system that blends a trained model with a website chat experience and a managed knowledge base. The setup supports training on data such as PDFs, website content, and even YouTube videos, then routes work between humans and AI: humans remain first-line support, while AI assists with tickets and emails to carry the heavy load—especially for small, bootstrapped teams.

For content creation, the stack leans heavily on image, video, and 3D generation. Replicate is used to run and fine-tune open-source models, including Stable Diffusion variants and Flux, with Flux described as a top image model and “Fox” highlighted as a leading option. Replicate also supports fine-tuning; the creator claims to have trained a LoRA model using images of himself to generate hyper-realistic thumbnails. For deeper control and custom pipelines, ComfyUI is used with node-based workflows and controlnets (depth maps and outlines), including cloud or local execution. For quick 3D-like environments, Depth is used to predict depth from a 2D image, turning it into a mesh that can be brought into After Effects.

Social distribution is managed with FeedHive, an AI-assisted scheduling tool spanning LinkedIn, YouTube, Instagram, TikTok, and more. It includes an AI assistant fine-tuned for content writing and a feature (V-dive) that predicts post performance and suggests edits to improve reach. FeedHive also supports automation inside AI agent workflows.

Finally, product development is accelerated with Cursor, a code editor with built-in AI features. Experienced engineers get predictive autocomplete and rapid edit suggestions, while non-coders can describe what they want in natural language and have Cursor help assemble an entire codebase. The overall message is that AI can run business functions in “10x mode,” but the practical advantage comes from combining tools into connected systems rather than using chatbots in isolation.

Cornell Notes

The core claim is that a SAS business can cut headcount dramatically by replacing routine work across operations, customer support, content creation, and product development with AI—while maintaining (and even improving) profitability. ChatGPT remains a central interface, but cost pressure pushes the use of OpenAI models through TypingMind. n8n provides the automation backbone, including agent setups that can autonomously choose tool actions instead of relying on complex hand-built flows. 8base handles support by training on company data and assisting with tickets and emails through a website chat and knowledge base. For content, Replicate, ComfyUI, and Depth help generate images and 3D-like assets, while FeedHive schedules and optimizes posts. Cursor speeds coding for both engineers and non-coders.

Why does the stack start with multiple AI chat options instead of one chatbot?

The workflow treats chat as a general-purpose interface, but different models fit different tasks. ChatGPT is described as the most useful for long, broad conversations. Gemini is singled out for maps/location-style questions (“where I can get something in my local area”). Claude is preferred by many founders, DeepSeek is noted for impressive recent results, and Llama is available as an open-source chat option. The practical twist is cost: ChatGPT’s $200/month plan is considered too expensive, so OpenAI models are accessed directly via TypingMind to get better value.

How does n8n change automation compared with simpler no-code tools?

n8n is positioned as more flexible than Zapier/Make because it supports advanced automation while allowing code when needed. It can also act as a backend prototyping tool by setting up API endpoints through drag-and-drop. The standout capability is agent integration: connect n8n to OpenAI, add memory and the tools the agent needs, then ask the agent to perform tasks. The agent figures out the best way to use the tools, reducing the need to manually wire complex conditional logic into the workflow.

What does 8base add that a fully custom “agent + RAG” setup might not?

8base is presented as a packaged AI support system that includes training and deployment components. It can be trained on business data like website content, PDFs, and even YouTube videos, then placed as a support chat on the website. It also manages a custom knowledge base and can assist with tickets and emails. The operating model is hybrid: humans remain first-line support, while AI handles a large share of the workload—aimed at small and bootstrapped teams that need fast, high-impact support automation.

How do Replicate, ComfyUI, and Depth divide up content production work?

Replicate is used for running and fine-tuning open-source diffusion models without managing infrastructure; it handles GPU provisioning and outputs results from prompts. ComfyUI is used when fine-grained control is needed: node-based pipelines, model chaining, and controlnets (depth maps and outlines) to guide image generation more precisely. Depth is used for a lower-effort 3D effect: it predicts depth from a 2D image, producing a mesh that can be imported into After Effects to create 3D-like environments without complex diffusion setups or advanced Blender work.

What role does FeedHive play beyond scheduling social posts?

FeedHive is described as an AI-powered social media management tool that schedules across major platforms (LinkedIn, YouTube, Instagram, TikTok, and more). It includes an AI assistant fine-tuned for content writing directly where posts are drafted. It also uses V-dive to predict how a post will perform and suggests edits to improve reach. It further supports automation inside AI agent workflows—so content creation can be integrated into the same agent-driven systems used elsewhere.

How does Cursor support both experienced developers and non-coders?

Cursor is framed as a VS Code-like editor with built-in AI tooling. For experienced engineers, it offers predictive autocomplete that anticipates where the next code change should go, enabling rapid “tap-tap-tap” edits. For people with no coding experience, it can be prompted in natural language to help assemble an entire codebase for web apps and mobile apps. The transcript also cautions that unrestricted AI coding without guardrails can get out of hand, implying no-code tools may be safer for true beginners.

Review Questions

  1. Which tool in the stack is presented as the automation backbone for building AI agents, and what specific capability reduces the need for complex conditional workflows?
  2. How do Replicate and ComfyUI differ in how they handle diffusion model execution and control?
  3. What hybrid support model does 8base use, and what kinds of data can it train on?

Key Points

  1. 1

    The business reports near-90% profit margins after shrinking a nine-person team to four by replacing many tasks with AI tools.

  2. 2

    ChatGPT remains a general-purpose AI chat option, but TypingMind is used to access OpenAI models directly to avoid a $200/month ChatGPT plan.

  3. 3

    n8n functions as the automation layer, enabling advanced workflows and agent setups that autonomously decide how to use tools.

  4. 4

    8base provides a packaged AI support system that trains on company data and assists with tickets and emails while humans stay first-line support.

  5. 5

    Replicate, ComfyUI, and Depth split content creation into infrastructure-free generation, fine-grained node-based control, and low-effort 3D-like environments.

  6. 6

    FeedHive combines scheduling with AI writing assistance and performance prediction (V-dive) to improve social reach.

  7. 7

    Cursor accelerates development for both engineers (predictive edits) and non-coders (natural-language codebase generation), with a caution about guardrails.

Highlights

Profit growth is paired with a major operational shift: headcount drops from nine to four while AI takes over core business functions.
n8n’s agent setup is described as a way to avoid handcrafting complex conditional automation flows by letting agents choose tool actions.
8base is positioned as a hybrid support system—AI assists with tickets and emails while humans remain first-line support.
A practical content pipeline emerges: Replicate for generation, ComfyUI for controlled diffusion pipelines, and Depth + After Effects for 3D-like scenes.
Cursor is framed as a productivity tool that can serve both coding experts and non-coders, though unrestricted AI coding can be risky without guardrails.

Topics

Mentioned

Get summaries like this for any content

Videos, articles, research papers — all summarized by AI and searchable in one place.

Start building your library