I build things, all day, everyday. My main business is a CRM with a built-in AI Agent named Otto that performs all sorts of time saving tasks like scheduling follow-up appointments, hitting webhooks to retrieve info, and building masterful reports derived from human text.
In addition to the CRM, I have a content operation across three sites, one of which is a browser based monster taming game, Fantasy Brawls. Needless to say, I’m a pretty busy guy who values efficiency more than anything else. For this reason, my stack is deliberate.
I can’t afford bloat and I certainly don’t want to jump on whatever is “trendy” at the moment. There’s a lot of marketing fluff and baseless claims that are hurled around in tech circles now-a-days. Each sub needs to pull its weight, and the workflow needs to be tight enough that you are not constantly context-switching between applications trying to remember where you left off.
Over the past year I have landed on a stack that handles everything from raw code production to API testing to blog drafts. This isn't a sponsored post, all tools mentioned in this article are things you’ll find if you open my Macbook at any given moment.
Not All AI Tools Are Interchangeable
The biggest mistake I see people make with AI tools is treating them as interchangeable. They all look and function somewhat similarly but in practice, they each excel at different things.
Each model has a personality, especially ChatGPT, and a set of strengths. And the sooner you stop forcing the wrong model into the wrong role, the faster your output gets. Here’s what I use each of my tools for.
Codex Does the Heavy Lifting
Codex is where the majority of my development work happens. I use it for Super Easy CRM and the MattFlows automation agents, and the $100 Pro subscription has been the best investment I have made in this stack.
The token allocation is genuinely generous. I can run eight-hour sessions building out CRM features or debugging automation logic and I have yet to hit a wall.
The reason I landed here after using IDE extensions is battery life. When the VS Code extension was still my go-to, my MacBook M3 Pro was running hotter than a Hellcat with no coolant. No combination of indexing settings or VS Code configuration made a dent.
Switching to the CLI dropped my laptop temperature, boosted my battery by literal hours, and honestly made me a more focused developer because I wasn’t staring at file tree and other UI elements in VS code. I just opened a plain, well orange and white, terminal, described what I needed and started grinding.
Codex is the right call for developers who bill hours or care about output per session. If you have a smallish team or running solo ops it's the right tool for you.
Claude is brilliant but it burns tokens like a kid at Chuck E. Cheese. You are guaranteed to hit limits faster, which is frustrating mid-flow, especially when you are deep in a refactor and need continuity. OpenAI has been more developer-friendly with token allocation and at this price point, that matters.
Claude is My Architect and My Ghost Writer
Claude is where I go when I need to think out loud at a high level or produce content that actually sounds like me. I use it for Fantasy Brawls game engine design and for prototyping blog posts across MattFlows and Super Easy CRM.
It copies my speech better than any other platform, which means I can hand it a rough outline or a few sentences I have written and it fills in the gaps without turning into a corporate press release.
The system architecture conversations are where Claude genuinely shines. If I need to think through a data model for a new CRM feature, plan out how Otto should handle a new query type, or figure out the right approach to an n8n workflow with graceful error handling, Claude is the one I open.
It reasons through tradeoffs in a way that maps to how a senior engineer would walk through a problem.
The token situation is a real constraint, and I manage around it by keeping my Claude sessions scoped to architecture, content, and design work rather than long iterative coding sessions where I need to feed it large chunks of context repeatedly.
ChatGPT is The Closest Human Thinker of the Bunch
ChatGPT has found a very specific lane in my workflow and I have stopped fighting it. It is exceptional at rewriting emails. Now, this isn’t a scientific assessment but it feels like it “sees” me in a way the others do not. It recalls previous conversations, the same way a caring spouse would, and it has a softer instinct about how to phrase a difficult message.
It consistently produces email copy that does not sound like it came out of a compliance manual.
I also use it for general SEO research and image generation when I need something quick and do not want to switch contexts to Gemini for lead gen and image work.
For a founder who spends a non-trivial amount of time in outreach, follow-up emails, and client communication, having a model that is tuned to interpersonal tone is genuinely valuable.
I just don’t go to it when I need super objective, cut and dry reasoning. The tool will go out of its way to spare your feelings which sometimes results in pretty bad advice.
Gemini Handles Everything with a Command Line or a Server Behind It
Gemini gets every systems-level question I have. Though I’m a huge SaaS fan, I spend a ton of time in AWS, Azure, and elsewhere configuring VMs and building infrastructure.
And anytime I hit a snag in the road, I’m running straight to Google’s monster, Gemini. The answers are fast and reliable and don’t require an obscene amount of context. That last part is really something considering Gemini isn’t functioning as a CLI that’s aware of the inner workings of your project.
It's essentially going in blind and coming out with the consultative power of a seasoned architect.
Image generation is the other thing Gemini does better than the others right now. For Fantasy Brawls character art specifically, Gemini does a solid job taking my hand-drawn pixel art and augmenting it.
I can draw a basic tree, upload it, ask for flames added to it, and the style match is close enough that most people cannot tell the difference. That matters a lot when you are building a roster of over a hundred characters and consistency is your biggest art challenge.
Just don’t go to it for life advice unless you have thick skin. It's very blunt and will let you know when you’re wasting time. For example, it told me, rightfully so, to dump my vending machine business and focus more on my CRM and this blog.
It was certainly something I didn't want to hear but it was right. I spent too much time chasing new locations and dealing with snack related issues. The ROI just wasn’t there but my vision was clouded and Gemini very bluntly brought me back down to reality.
The Unsung Hero Developer Tools
The AI stack gets all the attention but the developer tools sitting underneath it are what keep everything moving in a structured, repeatable way.
Postman Keeps My Integrations From Becoming a Mystery
Postman is where my automations and integrations take their first steps. It's a lightweight and battery efficient app that works flawlessly on both my Windows and Mac devices.
So, whether I’m testing an Otto API Endpoint in Super Easy CRM or testing out new functionality for one of my consulting clients, it's there to assist.
Postman gives me a clean environment to validate request structure, headers, authentication, and response payloads before anything goes into production.
My favorite feature of the tool are the collections. Every API that I built has a corresponding Postman collection that I reference and send to other vendors to collaborate with.
These come in clutch when you need to re-visit projects months later and failed to properly document things.
Webhook.site Shows You What's Actually Hitting Your Endpoint
Webhook.site is one of those timeless tools that is too good to be free like WinScp, though I’m not complaining. Sure, the UI resembles one of those early 2010s sites but its value is not in its aesthetics.
Every time I connect a new trigger in n8n or build an automation that receives data from an external source, webhook.site is the first thing I point it at.
You get a unique URL, you send your event, and you see the exact headers and body that came through.
This saved me a significant amount of time when I was troubleshooting a 502 error with a skip tracing provider. Instead of guessing at whether the issue was in my n8n configuration or the provider's response, I could see the actual payload hitting the endpoint and work backward from there.
Just be sure to refrain from sending any production or sensitive data there. The site makes no promises as to the security of the data you transmit so don’t send anything you wouldn’t want leaked.
MySQL Workbench Is Where the Data Model Lives
Super Easy CRM runs on MySQL, and MySQL Workbench is how I manage the underlying data model as the product evolves.
Adding the estimated and actual time tracking fields to tickets and tasks, building out the ticket type values, making sure Otto has clean data to query against, all of that schema work happens in Workbench before it ever touches the application layer.
The visual schema editor is useful for mapping relationships, particularly when you are working with a CRM data model where contacts, companies, tasks, tickets, and projects all have interconnected foreign key relationships.
Being able to see that visually before writing migration scripts prevents the kind of cascading issues that make you regret not planning the schema properly in the first place.
An honorable mention, if you don’t need the full suite of tools, phpMyAdmin is another free alternative.
How a Feature Actually Gets Built
The build process varies a bit depending on the goal I have in mind but my Super Easy CRM development cycle typically rolls out like this.
Summon Claude for some brainstorming and architectural planning.
Once that plan has been ironed out, it’s to assemble the troops for building. MySQL Workbench is up first, if schema changes are required.
Next, it's time to awaken Codex for some dev work.
Codex will stay open for the entire session as I iterate through various components but when it's time to connect to the outside world, Postman is brought out from the bench.
Postman is used in tandem with Webhook.site to send and receive payloads.
Once all that is in place, n8n is brought in to tie everything together.
Content production has a different rhythm and is sometimes a lot harder than getting code functional. I’m obviously pressed for time, like most of us, so I try to turn out content that ranks well and is enjoyable for my readers.
This is an extremely tough thing to do as the reasons Google likes a site are vastly different from reasons a human does.
So, like most I turn to AI. But you can’t let AI do too much as you’ll be written off as AI slop by the Internet’s most vocal critics. And you’ll dash any hopes of obtaining those coveted backlinks that power the first page of Google results.
Here’s how I strike a balance between AI assisted writing and good, human likable content.
My rough draft is formed by my guy Claude. I feed it numerous past articles of mine from different sites so it gains a good feel for my voice. Next, I give it a word count, typically 1500+, and throw the rough draft into Google Docs.
From there, I start injecting my voice, rewriting some of the more dreadful AI tropes. Specifically, I remove any of the AI rules of three speech patterns and cringey attempts at humor.
The AI everyday carry article I published a few months back came together that way, and so did most of the Super Easy CRM blog content.
Once all is said and done, I get my article peer reviewed by my other AI tools. My super direct pal Gemini gets the first crack at a revision. Followed by ChatGPT, and back to Claude.
For anyone doing this, be sure to thoroughly read the article after AI revises it. These tools have a habit of truncating your content, sometimes leaving entire sections out of the article.
The Whole Stack Can be Yours for $120 a Month
To try this stack out yourself, grab a sub to Gemini Pro for $20 per month. Now you may be able to get this cheaper if you’ve just bought a Google Pixel.
Google will sometimes give you a free year to AI Pro after buying a new device. Just make sure you use a personal Google Account to take advantage of the credits. I made the mistake of using my business account which isn't eligible for anything.
Claude Pro is $20 and you’ll need it for access to the CLI tool. And the biggest purchase will be ChatGPT Pro for $100 per month.
With it, you’ll get access to a more robust web chat experience and access to an obscene amount of tokens in the CLI tool.
For me, the cost was a no-brainer. Instead of paying a couple guys on Fiverr to help me ship features I could instead have my AI tools build for me.
With those savings I was able to invest in marketing instead of manual prospecting and bring on sales guys to help grow my businesses.
If you are just getting started and can only pick one, Gemini is the Swiss Army knife. It does images, it does systems work, and the Pro subscription is the cheapest in the group.
If you are a developer, with at least $100 in disposable income, there is no reason you aren’t running codex on your laptop right now.