Subscribe | Free AI | CRO Audit | Cool Tools

Is AI Going Too Fast?

We built our first AI product, Ochatbot, in 2017. Generative AI was only used in a few large companies and governments at that time. Once OpenAI released ChatGPT to the public, the race was on for companies. The technological advances are now being measured every quarter, and in some cases, every month, increasing faster than Moore’s Law. It is like seeing the release of personal computers on steroids. The engineering geek part of me is in awe of living in this historic time.

However, I do have some concerns about people not understanding the limits of AI. To be fair, it is hard to understand the limits when they change every couple of months. Here are some examples:

• Vibe coding an application seems straightforward, especially the look and feel of the app, the functions, etc. What is not always understood is that AI is not omniscient. It does not know what companies do not publish, particularly trade secrets. In the end, you have a program that has 90% of the basic ingredients. Don’t assume AI is right.

• Security, compliance, and hardware. It goes without saying that you need to be at industry standards when it comes to cybersecurity. Some of these tasks are on the hardware side, some in the code. You need to know how to ask and test them.

• Agents building code while other agents test what’s built. There is something odd about this. Why can’t the agent build it right the first time? To me, this just shows that agentic technology is still developing. Our experience has been that you need to assume Murphy’s law is in play and don’t take a logical approach to what you expect to be changed in your code.

• A good product works well for the main use case and handles all the edge cases. It takes years for a mature product to handle all the edge cases. Test the edge cases.  Watch out! If your vibe-coded app missed an edge case, your AI-generated tests probably also missed that edge case.

Don’t get me wrong, I have seen CRMs, simple chatbots, APIs, and a list of tasks that manipulate data, all created by non-programmers with Agentic AI. Most of these are used internally in a company or for a single use case. This is very impressive. All I ask is that before pushing the product to the public, they have an engineer review and test it. Ironically, I would not be surprised if, in twelve months, the ability to create code and processes agenticaly will have a significantly reduced error rate. But I am not convinced it will replace software with a long history. If anything, Agantic AI is forcing SaaS companies and developers to raise the bar.

News This Week

• LLMs are too important to be left to big tech. There is a gap between frontier models and models that can run on your device, but local models improve each day. (Link)

• Forge is a system for enterprises to build frontier-grade AI models grounded in their proprietary knowledge. Forge bridges the gap between generic AI and enterprise-specific needs. Instead of relying on broad, public data, organizations can train models that understand their internal context embedded within systems, workflows, and policies, aligning AI with their unique operations. (Link)

• OpenAI Scraps Sora Video Platform Months After Launch. The app, once envisioned as a potential social network, allowed people to insert themselves into various pop-culture scenes (Link) WSJ

• Notes from the SaaS Funeral - Reid Hoffman - Now that you can walk up to Claude Code, Microsoft Copilot, or OpenAI Codex and say “make me a CRM system,” why would anyone pay for enterprise software again? (Link)

• I wanted to build a vertical SaaS for pest control. I took a technician job instead. (Link)

• OpenClaw: Our Comprehensive Guide for Beginners. A super-smart, always-on agent who automates the tasks you dread—and moves your biggest ideas forward (Link)

• Why CPC keeps rising – and what to do (Link)

• Google March 2026 Spam Update Unleashed (Link) And in case you missed it, Is Google Hitting Self-Promotional Listicles Hard Again (Link)

• Google is testing a new way of showing citations in AI Overviews. Honestly, this feels like a throwback to the days of “SGE.” (Link)

• How to write for AI search: A playbook for machine-readable content. Learn how to structure clear, information-rich content that LLMs can extract, interpret, and cite in AI-driven search. (Link)

• Schema won’t guarantee citations, but it helps AI understand entities. Here’s how to use structured data for clarity and cleaner extraction. For AI, three elements matter the most: Entity definition: Which brands, authors, services, or SKUs exist on the page. Attribute clarity: Which properties belong to which entity (e.g., prices, availability, ratings, job titles).​ Entity relationships: How entities connect (e.g., offeredBy, worksFor, authoredBy, and sameAs schema tags).​ (Link)

Cool Tools

AI tools for sales

• Ensu - Ente's Local LLM app (Link)

• Warp - Analytics tool (Link)

• Illuiminarty - Detection of AI-generated images (Link)

• CensysGPT - Security analysis (Link)