AI for Building an AI-Native Business Webinar April 2nd

This session is a Q1 check-in rather than a single business use case: a tour of where the technology actually stands, what is useful and what is fluff, and how I am rebuilding my own business from the ground up to be as AI-native as possible (1:56). It is for anyone who wants to see a real working stack, with live builds included, not a slideware version of one. The core tools are Claude Code and Gemini, plus Apify, Unipile, and Modal for the advanced use cases.

The software factory concept (8:28)

I opened with a blog post by my friend Oost called The Software Factory, because I believe it describes how a lot of us will work in the near future. He defines two things: a factory and a primitive. A primitive is the most foundational building block of software, the Lego piece. Nothing is smaller than a Lego piece.

In the previous paradigm you had two routes. Either you build a SaaS, which is highly scalable but not really customizable because you have to make cuts to serve as many people as possible. Or you do consulting and custom development, which is incredibly customized to one client's workflow but impossible to replicate. Thanks to AI, you can now keep a library of Lego pieces, modules for social tracking, lead qualification, reporting to WhatsApp or Slack or email, and just ask the AI to stick them together. That gives you something both extremely scalable and extremely customizable. The catch: it only works if your foundations are right, and that is the hard part I am working through myself right now.

Claude Code: the center of command (12:24)

I polled the room and, to my surprise, a good chunk of people were already on Claude Code, several on the Max plan. My whole setup is built around one principle: Claude Code is always my center of command. You can substitute any other AI coding agent, Codex or OpenCode work too, but there is one interface for the entire business.

That interface connects to my knowledge base, and the knowledge base connects to all the external sources: emails, content, call transcripts, Google Drive, calendar. Just those five tools around one knowledge base show you how quickly the ecosystem gets messy, which is exactly why the structure matters.

Your knowledge base is everything (15:08)

The most underrated element of this whole architecture is the knowledge base. It is your central brain, the single source of truth for your entire business operations: onboarding and offboarding documents, client transcripts, your CRM, your testimonials, and your content if you produce any. When someone asked what a knowledge base actually is, my answer was that it is fancy words for something super simple. Mine is a Notion workspace. Yours could be Google Drive documents, Obsidian, or even CSV files inside your Claude Code workspace. It is not about technology.

Making it AI-ready is the real work (17:00). My approach: connect Claude Code to your Notion or Drive, have it map out the totality of the workspace, pull the schemas of every database, how they are built and how they connect, then reorganize everything in the most logical way and do a deep dive together to find the missing connections. AI is absolutely brilliant at categorizing data. But the first pass is always human, and this categorization is where you spend 80 to 90 percent of your time. When it is done, the rest is easy.

I showed my own workspace live (20:12): the CRM I am rebuilding, a competitors dashboard that will feed an automated tracking system, and databases for every piece of content I have ever produced. The honest confession behind all of it: the day before this webinar I decided to burn my business down and rebuild it AI-native, because it is easier to start clean than to bolt AI onto broken structures.

The great debate: n8n and Make.com versus Claude Code (22:12)

Everyone is saying n8n is dead and Make.com is dead. I disagree, and it is not even an opinion, I am convinced there is a time and place for everything. Humans are visual beings. If you work with a team, an n8n or Make workflow can be shared and documented so anyone can change a prompt, an API key, a model, or a database. That is genuinely valuable.

Claude Code is not visual. It is better for solo operators, which most of us are. And Claude Code by itself cannot schedule and run things the way n8n or Make can, which is why you pair it with a deployment service. My pick is Modal. When someone later asked how to choose for a client (56:42), my rule of thumb was: if the client is not willing to learn AI and does not want to debug things themselves, go Claude Code plus Modal and run it for them.

The stack: Claude Code, Unipile, Modal (26:18)

Three Lego blocks power most of what I showed. Unipile is a service that connects your social accounts via API: LinkedIn, WhatsApp, Instagram, X, Telegram, everything except Microsoft accounts. It gives you access to data that is otherwise locked inside your LinkedIn account, and it also handles outreach actions like connection requests and messages.

Modal (29:05) is pretty much a service that lets you put Python scripts on the web and run them whenever you want. The free tier is extremely generous: I do not even spend the five dollars of monthly free credit, and they add around thirty dollars if you register a credit card. Realistically you may never pay for it. At the bottom of my dashboard you can see my two standing deployments, LinkedIn Metric Sync and YouTube Metric Sync, which run every morning and refresh again at 7 PM, keeping my Notion content databases up to date.

This is also exactly how you deploy automations for clients: a thirty-minute onboarding call to create the accounts together, build the workflow with Claude Code against their Unipile and Modal accounts, deploy, then disconnect your API access and hand it over (32:49).

Live build: LinkedIn post sync to Notion (31:33)

To prove the point, I built one live. I asked Claude Code to fetch all my LinkedIn posts through Unipile: 113 posts came back with engagement metrics included. Claude then spotted that 47 of the 113 were reposts. So I asked it to create a Modal deployment that captures that repost information and saves it into a new Notion database it creates itself. A few minutes later, the LinkedIn all post sync deployment appeared in Modal, fully wired to Notion. That whole loop, fetch, analyze, deploy, sync, is something you can genuinely build in fifteen minutes with a Modal account, a Unipile account, and Claude Code.

Gemini for video analysis (37:36)

I highly recommend marketers start looking at Gemini. Via API, Gemini analyzes videos properly. It is not a transcription tool bolted on, you are literally just using Gemini, and it is that good. I had Claude Code pull one of the 45 videos from my YouTube Notion database, Get your money's worth from ChatGPT, and analyze it with Gemini through the API: full transcript with very accurate timestamps, duration, everything, rendered into an HTML analysis page on my desktop.

The use case for content systems is obvious: point it at a database of one or two hundred videos and ask it to tag when, say, a woman is talking about physics. No complex video pipeline, just Gemini. And my favorite party trick (48:34): I filmed myself writing equations on paper, showed each one to the camera, and Gemini solved them. Not simple ones, master-level equations. Video analysis is about to jump through the roof.

The broader pattern (53:27) is that my Claude Code holds API keys for all the major LLM providers, plus access to my Gmail and other services. When I tell Claude Code to use Gemini, it calls the Gemini API. One single interface for your whole business, with each model doing what it is best at, and right now Gemini has the edge on documents, video, and audio.

Full applications on Modal and the WhatsApp calculator (42:15)

Modal is not just for cron scripts, you can deploy entire applications with it, which honestly I only discovered when Claude Code did it. I demoed a tool it built for me: a calculator that estimates what a WhatsApp marketing campaign will cost, for example 50,000 messages toward France (50:02). Then I asked Claude Code to rebuild it so the report goes out via LinkedIn instead of WhatsApp: it generated a PDF quote and delivered it to David from the audience as a LinkedIn message, live. That version took about five minutes. Think about lead magnets like this: someone drops their LinkedIn profile or phone number, and a personalized report lands in their inbox. When the demo was done, I told Claude Code to destroy the app and shut down the deployment. Lego pieces, assembled and disassembled at will.

On data security (44:01): for Claude Code you can contact the Anthropic sales team about zero data retention and turn off the setting that shares data to improve Claude. Google says they do not train on your Gemini API data; I mostly work with public information anyway. For genuinely sensitive clients, local models on your own machine are the fallback.

The LLM comparison dashboard (54:38)

I also showed a dashboard I built by connecting the API documentation of all the major LLM providers. It breaks down cost for the flagship and cheapest models of each, side by side with their distinctive advantages: Gemini, for example, is the only one that is truly multimodal, takes any file type, has a one million token context window, and offers a free tier. If you want it, message me on LinkedIn.

Q&A highlights (56:42)

• How to learn Claude Code: sit down and ask it questions. Doing is the best way to learn, and you will learn what is relevant to you instead of fluff. I have a tutorial out, and Nick has one that runs about four hours.
• Who I target and pricing: I am deliberately broad right now, mostly small and medium businesses with some enterprise. Our average ticket is around three thousand euros, we are setting the baseline at two to three thousand, and our biggest client was eighteen thousand euros at enterprise level.
• Do these webinars generate leads? Not directly, but they generate social proof, and that compounds.
• Getting email data into Notion regularly: an API connection to Gmail through Claude Code, or Unipile, then either a scheduled task on your own machine or a Modal deployment that runs it for you.
• Infographics for LinkedIn: Claude Code generates HTML pages that get screenshotted into finished visuals.
• Claude Cowork: if you already use Claude Code, you do not need it. It is a lighter version of the same thing.

Demo: Night Agent, Prospector, and Observer (1:05:41)

To close, my three bread and butter LinkedIn workflows, all built with Claude Code and Apify, nothing else.

The Night Agent checks who has viewed your LinkedIn profile, scrapes those profiles, qualifies them against your ICP, and surfaces the qualified leads. I recently improved it to track people who repeatedly view my profile. Under the hood it is an MCP connection to Playwright that scrolls and scrapes, plus Apify, and it does require LinkedIn Premium.

The Prospector fetches posts on a given niche across LinkedIn and classifies the posters against my ICP so I know who to engage with. The Observer analyzes the performance and engagement of all my LinkedIn content: a database of every engager, which posts they interacted with, profile data, and insights on my best-performing content and top engagers. Neither of these two needs Premium, and Prospector is honestly the easiest one to build.

I run these sessions every single Thursday, beyond the monthly MakerSchool slot. Catch the next one or browse the replays on the webinars page.