Agentic AI has huge potential to revolutionize our perception and usage of Large Language Models. The decision-making powers of an AI Agent incorporated into the automation of the Agentic systems are an important step in providing value statements to AI. However, there is a real problem with Agentic AI: Most Agentic AI programs are fucking trash.
I do not say this lightly: I am an Agentic AI programmer myself. That doesn’t change the fact that most Agentic AI programmers have three major failings:
- They don’t know the difference between AI Automation and Agentic AI
- They are too dependent on Vibe Coding
- They have no business experience
What is Agentic AI?
There are two major types of AI usage in this space: AI automation and Agentic AI. Here is a description of each using your vehicle’s need for an oil change as an example:
AI Automation – Rigid, Time-Based Scheduling
This method sticks to a pre-set calendar, like a reminder app:
- Fixed Intervals: Triggers service alerts based on elapsed time (e.g., every six months), regardless of actual driving habits.
- No Context: Ignores mileage changes, trip plans, or sensor feedback.
Result: “Oil Change Scheduled: Aug 15, 2025” – even if your oil’s fine or you’ve just driven 2,000 extra miles.
Agentic AI – Adaptive, Context-Aware Scheduling
This system thinks like a proactive mechanic with access to your entire driving life:
- Mileage Monitoring: Tracks actual driving distance, not just time, so frequent drivers get earlier alerts.
- Outlook Integration: Analyzes upcoming trips or meetings from your calendar to predict upcoming mileage surges.
- Sensor Input: Continuously checks oil levels and quality, adjusting recommendations dynamically.
Result: “Oil Change Recommended: Aug 10, 2025” – predicted ahead of time because you’ve got a road trip planned and your oil is trending low.
AI Automation versus Agentic AI
There are two ways in which the difference between AI Automation and Agentic AI leads to trashy implementations: coders don’t know the difference, or they do and sell it as Agentic AI anyway. The first generally occurs from new programmers who have learned from YouTube how to vibe code or use mostly AI to write code for them, and the second occurs when coders want to charge more for their work despite the actual output being fairly simple.
A true Agentic AI system can start at tens of thousands of dollars, and takes weeks or months to create. AI Automation is much easier to create, often simply pushing data through APIs that take hours to program, not days or weeks. However, charging for hours worth of work as if it took days or weeks is very lucrative, especially when that programmer can reuse the same code over and over. Simply put, building AI Automation and charging for it like it’s Agentic AI is great for profit margins.
It is not, however, good for your business’s bottom line. There are use cases for both AI Automation and Agentic AI, but paying for one to get the other is costly. If you only need AI Automation, pay for AI Automation.
There is a sinister element to this that requires you to be educated as an employee of a company looking to buy these kinds of accelerators. I have seen the companies selling AI Automations as Agentic AI: They are not just young people just starting their career in vibe coding. Selling you an automation that requires minutes to configure as a program that requires hundreds of hours to create is commonplace right now, as buyers of these offerings aren’t educated enough to tell the difference.
The Dependancy on Vibe Coding
Vibe coding is a form of programming that uses AI to create code based on natural language prompts. It simplifies the process of coding what used to take months into hours. n8n, Lovable, and Replit all use vibe coding as the basis for their software. Vibe coding allows for people with no coding background to quickly create code, and is all the rage in programming circles right now.
The problem with vibe coding is it allows people with no programming experience to pretend they have coding experience. When they run into trouble with their code, they have no idea how to fix it, except to send another clarifying prompt to the LLM. LLMs are already a black box to the majority of us, meaning we have no idea how they actually function. Vibe coding allows you to build black boxes around those black boxes, with the people performing the vibe coding having no idea what they are actually creating and how it works.
This is how trash AI solutions are created. Somebody with no programming experience, prompting an AI system to create code for them, is 100% dependent on the AI to get it right. They can’t troubleshoot, and therefor,e they can’t support their own platform.
This isn’t to say vibe coding is bad. It’s a force multiplier in how code is written. However, it should not be a substitute for at least some programming experience. Our partners at Patchwork AI have the experience and the tools to create usable, adaptable, sustainable Agentic AI agents and AI Automations. Choosing the right vibe coders is just as important as writing the right requirements to avoid paying for trash AI solutions.
Business Experience is Required
A fair amount of the Agentic AI being created today is being produced by programmers with no business experience. They have never been through a discovery project, don’t know how to write business requirements, have never performed UAT, Unit, or Smoke testing, and are assuming that their vibe coding skills can make up for those deficiencies.
Let me be perfectly clear: That’s bullshit.
In order to create a sustainable, adaptable, supportable product, there MUST be discovery. There MUST be requirements. There MUST be testing. There is nowhere else in your business that you would trust a lack of experience in a specific set of topics within the domain of your business to run a project like this. So why do you do it with AI?
It’s actually pretty simple: People are scared of what they don’t know about AI, and default to anyone with any explanation that seems plausible without any amount of due diligence. Someone walking in telling you how they’ll vibe code you an AI agent that will automate a task sounds great, but doesn’t mean that they know how your business operates, your go-to-market strategy, your branding, your teams, your tech stack, or your support requirements.
Having vibe coded myself, I can tell you that it is easy to get caught in the belief that you can skip the elements of a normal project, ignore project management, and release MVP-level code with no plan on how to move to more stable, supportable releases. If programmers don’t understand abstraction, polymorphism, and encapsulation, their code, regardless of what vibes they had while coding it, will fail over time.
The Fucking Summary
If it isn’t obvious by now, educating yourself about what Agentic AI actually is and treating these projects like you would any other project is vital to ensuring your investment in AI pays dividends. With tariffs testing profit margins and inflation slowing spending, investing in poor technology is not a recipe for success. Ask questions, run projects the way you know how to run projects, and make sure you have a trusted partner that understands your business. These are the keys to AI success.