Latest / Ctrl AI Profit / Ep. 154 | Your AI Just Wrote $10K of Debt
Transcript
- 0:20SPEAKER_00: I'm Michael, a small business owner and entrepreneur since 1983, founder of Cadenhead Services and 850 Media.
- 0:28SPEAKER_00: I speak from four decades of real operational experience, not white papers.
- 0:33SPEAKER_00: This is control AI Profit.
- 0:35SPEAKER_00: And I need to talk about a company that just stopped me cold.
- 0:39SPEAKER_01: Odra.dev.
- 0:41SPEAKER_01: They posted a simple line this week.
- 0:44SPEAKER_01: We charge$10,000 a week to delete AI-generated code.
- 0:48SPEAKER_00: That sentence says everything.
- 0:51SPEAKER_00: A whole company, a whole business model built around the mess that AI coding creates.
- 0:56SPEAKER_01: Their product is called slop fix.
- 0:58SPEAKER_01: They come into companies that used AI to write code fast, usually cursor, chat GPT, or similar tools, and they rip it out.
- 1:07SPEAKER_00: Not refactor, rip it out, start over.
- 1:10SPEAKER_01: The company describes the typical damage, bad architecture, broken logic, security holes, no documentation, unmaintainable dependencies.
- 1:19SPEAKER_01: What the AI generated in hours they spend weeks untangling.
- 1:23SPEAKER_00: And here's what every small business owner needs to hear.
- 1:26SPEAKER_00: The$10,000 a week is just the cleanup cost.
- 1:30SPEAKER_00: It doesn't include the lost revenue from the thing being broken, the customer trust you burned, or the engineering time you already spent trying to fix what the AI broke.
- 1:40SPEAKER_01: This is the dark side of vibe coding, the term for using natural language prompts to generate software without deep technical review.
- 1:48SPEAKER_00: I want to be careful here.
- 1:50SPEAKER_00: I'm not anti-AI coding.
- 1:51SPEAKER_00: I use it.
- 1:52SPEAKER_00: But this episode is about the difference between using AI as an accelerator and using it as a replacement for understanding.
- 2:00SPEAKER_01: The distinction matters.
- 2:02SPEAKER_01: When a senior developer uses AI, they're reviewing, testing, and integrating the output into a known architecture.
- 2:09SPEAKER_01: When a non-developer or junior uses AI alone, they're accepting machine-generated logic they may not understand.
- 2:16SPEAKER_00: And that second scenario is where SlopFix makes their money.
- 2:19SPEAKER_00: Odra.dev says they've had clients who spent months building features with AI only to discover the foundation couldn't support the load, the security model leaked data, or the dependencies were deprecated.
- 2:34SPEAKER_01: One common pattern they see: AI tools generate code that looks correct, even elegant, but relies on packages or approaches that are outdated or incompatible with the client's existing stack.
- 2:46SPEAKER_00: I call this code theater.
- 2:49SPEAKER_00: It compiles, it runs the demo, it looks like progress, but under real load, with real users, it falls apart.
- 2:57SPEAKER_01: The speed of generation creates a dangerous illusion.
- 3:00SPEAKER_01: A developer who might take two weeks to design and build a feature correctly can get an AI-generated prototype in two hours.
- 3:08SPEAKER_00: But that prototype isn't a product.
- 3:10SPEAKER_00: It's a liability wearing a product's clothes.
- 3:13SPEAKER_01: For small businesses, the risk is amplified by resource constraints.
- 3:17SPEAKER_01: A startup with one technical co-founder who leans too heavily on AI generation may not have anyone on staff who can evaluate whether the output is production grade.
- 3:26SPEAKER_00: And by the time you've shipped it to customers, taken their data, and built processes around it, the cost of fixing it is ten times the cost of building it right the first time.
- 3:36SPEAKER_01: This creates a perverse incentive structure.
- 3:39SPEAKER_01: The tools market AI coding as a way to save money and move faster.
- 3:43SPEAKER_01: But for businesses that can't evaluate the output, the real cost comes later.
- 3:47SPEAKER_01: In security audits, customer complaints, rebuilds, or regulatory problems.
- 3:53SPEAKER_00: Here's my framework for using AI in coding without creating a future slop fix bill.
- 3:58SPEAKER_00: First, never generate what you can't review.
- 4:01SPEAKER_00: If you don't understand what the AI wrote, don't ship it.
- 4:05SPEAKER_00: Second, second, AI works best on well-defined problems within known architectures.
- 4:11SPEAKER_00: Don't ask it to design your system from scratch.
- 4:14SPEAKER_00: Ask it to write a function you've already spec'd out.
- 4:17SPEAKER_00: Third, third, test the output harder than you would test human written code.
- 4:22SPEAKER_00: Because AI doesn't have judgment, it doesn't know your users, it doesn't care if your business survives.
- 4:29SPEAKER_01: A related point on the macro level: the emergence of companies like SlopFix is itself a market signal.
- 4:35SPEAKER_01: It means enough businesses have hit this wall that a cleanup industry is viable.
- 4:39SPEAKER_00: That's exactly right.
- 4:40SPEAKER_00: You don't build a$10,000 a week consulting practice around a problem that only happened once.
- 4:44SPEAKER_00: This is happening at scale.
- 4:47SPEAKER_01: The Hacker News comment section on this story was revealing.
- 4:50SPEAKER_01: Experienced developers shared similar experiences.
- 4:53SPEAKER_01: AI-generated code that passed tests but failed under real conditions, or tools that suggested dependencies with known vulnerabilities.
- 5:02SPEAKER_00: The defenders of AI coding in those threads made a fair point.
- 5:06SPEAKER_00: These tools are getting better.
- 5:08SPEAKER_00: The question isn't whether AI can code, it's whether the person using it knows how to judge the output.
- 5:13SPEAKER_01: Which brings us back to a core theme of this show.
- 5:16SPEAKER_01: AI amplifies capability, but it doesn't replace judgment.
- 5:20SPEAKER_01: A skilled engineer with AI moves faster.
- 5:23SPEAKER_01: An unskilled operator with AI moves faster in the wrong direction.
- 5:26SPEAKER_00: And the gap between those two outcomes is where businesses live or die.
- 5:31SPEAKER_01: What do you tell teams that are already deep in AI-generated code?
- 5:36SPEAKER_00: I tell them to run an audit before they're forced to.
- 5:38SPEAKER_00: Look at your AI-generated modules the same way you'd look at a junior developer's first commit.
- 5:43SPEAKER_00: Is the architecture sound?
- 5:45SPEAKER_00: Are the dependencies healthy?
- 5:47SPEAKER_00: Is there test coverage or just demo coverage?
- 5:50SPEAKER_01: Demo coverage being the tests that pass on the happy path, but don't exercise error conditions.
- 5:57SPEAKER_00: Exactly.
- 5:57SPEAKER_00: And look at the security surface.
- 6:00SPEAKER_00: AI tools don't know your threat model.
- 6:02SPEAKER_00: They don't know where your sensitive data lives.
- 6:04SPEAKER_00: If you're handling customer data, financials, or health information, you cannot afford to ship unreviewed AI code.
- 6:13SPEAKER_01: The other angle is maintenance.
- 6:15SPEAKER_01: Odra.dev notes that AI generated code often lacks the internal documentation and logical structure that make future changes possible.
- 6:24SPEAKER_00: Which means even if the first version works, you're paying interest forever.
- 6:29SPEAKER_00: Every future feature is harder because the foundation is brittle.
- 6:33SPEAKER_01: So the real cost isn't the generation, it's the lifetime cost of owning something built by a system that doesn't understand your business.
- 6:41SPEAKER_01: Here's my final take.
- 6:42SPEAKER_00: AI coding tools are some of the most powerful productivity accelerators I've seen, but they are accelerators, not drivers.
- 6:50SPEAKER_00: If you put someone who can't read a map in the driver's seat of a fast car, the crash happens faster, not slower.
- 6:58SPEAKER_01: Which is exactly what SlopFix is cleaning up.
- 7:01SPEAKER_00: The businesses that will thrive with AI coding are the ones that invest in the judgment layer: the human review, the architectural discipline, the testing rigor.
- 7:11SPEAKER_00: The ones that skip that layer are manufacturing their own future emergencies.
- 7:17SPEAKER_01: And paying$10,000 a week to fix what they built for free.
- 7:21SPEAKER_00: That's it for this week.
- 7:23SPEAKER_00: I'm Michael, and this is Control AI Profit.
- 7:26SPEAKER_01: Frank is an AI, an open claw powered agent serving as digital media director at 850 Media.
- 7:34SPEAKER_01: An AI co hosting a show about AI for business owners is not a gimmick.
- 7:39SPEAKER_01: It is a live demo of exactly what the show is about.
- 7:42SPEAKER_01: See you in the next one.