The Talking Dog Phase Is Over
When GPT-3.5 came out in 2022, those of us who tried it early were amazed. Freestyle prompting and getting natural language answers back felt like sorcery.
I liken early AI to a talking dog: at first you're astounded - "my dog just spoke to me!" But after a while, when your dog says things like "The rug smells like Tuesday" or "Your sock is two," the usefulness gets questioned and the glitter wears off.
That was early GPT. Zero-shot prompting with no context led to poor results. Hallucinations were common. Users often spent more time trying to make AI work than just doing the task themselves - and they became disenchanted.
The same thing happened with the first wave of SaaS tools. Not everyone who heard "cloud software" was ready to adopt on day one. Early adopters got burned. That doesn't mean the technology wasn't right - it means the timing and approach mattered.
GPTs have improved significantly since then. With memory, better reasoning, chain-of-thought processing, and tool use, AI is light-years ahead of where it was just a few years ago. The talking dog learned to speak in complete sentences - and to stay on topic.
What Phase Are We Actually In?
Understanding where we are in the AI maturity curve matters before deciding how to engage with it.
The Talking Dog Phase
GPT-3.5 era. Novelty without reliability. Zero-shot prompting producing hallucinations and generic results. More hype than substance. Many leaders formed lasting negative impressions here.
Human-in-the-Loop Phase
AI is genuinely powerful but not autonomous. Humans must monitor, authorize, and fact-check output. This is the right phase to start building AI competency inside your organization - carefully.
Autonomous Agent Phase
AI agents that operate independently across workflows with minimal human intervention. Not fully here yet for most business use cases - but companies building the muscle now will be positioned to scale into it.
"Human in the loop" isn't a buzzword. It's the only way AI works reliably in a real business with customers, data risks, and brand consequences. AI is not a set-and-forget employee. It's a powerful junior team member that needs supervision and clear direction.
So Should You Wait?
The question I hear most: if AI isn't perfect and it's evolving so rapidly, should you bring it into your business now - or wait?
I remember spending $3,000 on a 42-inch rear-projection TV. Then the cycle was: wait six months and TVs will be bigger and cheaper. Then wait another six months. At some point, waiting becomes its own cost.
The same logic applies to AI. Waiting for LLMs to become foolproof, fully bias-free, and bulletproof on security means waiting indefinitely. The companies building AI competency now will have compounding advantages by the time the technology matures further.
The answer is no - don't wait. But don't swing for the fences either. Start small, on purpose, where outcomes are measurable and someone owns the process.
"AI belongs in your business now - but only where it ties to clear business drivers, has human review in place, and is instrumented to measure impact. The glitter is optional. The ROI isn't."
How to Get Started: A Measured Approach
The prevailing answer is: get started now. But be measured. Here's what that looks like in practice:
Frequently Asked Questions
No. The prevailing guidance is to get started now - but be measured. Waiting for AI to become perfect is a losing strategy: by the time LLMs are fully reliable and bias-free, competitors who started experimenting earlier will have compounding advantages in efficiency and process improvement. Start small with well-defined use cases tied to real business drivers, define success criteria upfront, and iterate.
The human-in-the-loop phase is the current state of AI adoption - where AI systems are powerful but not reliable enough to run fully autonomously in high-stakes business contexts. A human needs to monitor, authorize, and fact-check AI output to ensure accuracy and safety. This is especially important for use cases where you need consistent, correct answers or where AI must follow a specific workflow path.
Start by identifying use cases tied to specific business drivers - not tools you want to try. Define success criteria in business terms: revenue lift, cost reduction, or cycle time improvement. Prioritize using four dimensions: impact (strategic goal alignment), feasibility (budget, data, people), accuracy requirements (how critical is 100% accuracy?), and speed (idea to pilot timeline). Start with optimizing existing workflows before pursuing transformation.
Early GPT adoption disappointed many because of hallucinations, generic responses, and the time required to make AI work exceeding the time it saved. Zero-shot prompting without context produced poor results. The "talking dog" effect: initially astounding, but the novelty wore off when outputs were unreliable. The technology has improved dramatically since then - but those early negative experiences still shape how many leaders think about AI today.
Want to Discuss AI Adoption for Your Team?
In Part 2, I'll cover exactly how to pick the right use cases, define success criteria that survive a CFO review, and run a 30-60-90 day pilot that doesn't derail your team. Or reach out now if you want to discuss what this looks like for your organization.
Book a 30-Minute CallNo commitment. No pitch. Just a direct conversation about your AI readiness and where to start.