Maslow's Hammer and the 2 AI Extremes

In many meetings I'm in, companies are choosing AI extremes right now: total adoption or total rejection.

This polarization echoes Abraham Maslow's 1966 observation: "If the only tool you have is a hammer, it is tempting to treat everything as if it were a nail." Maslow described cognitive bias in problem-solving, where people default to familiar tools regardless of appropriateness. We're watching this with AI adoption at enterprise scale.

The Tool Selection Problem

Maslow's Hammer illustrates how we often force-fit solutions based on available tools rather than addressing the actual problems. The concept hits business strategy, technology adoption, and organizational change.

Maslow's framework is popping up specifically because AI is such a versatile tool. There are many ways it can be used and in many contexts, from automation to simple calendar organization. A tool like that is far beyond a hammer. It's duct tape.

When consultants discovered Six Sigma, every problem became a process optimization opportunity. When companies discovered digital transformation, every challenge required an app.

People are taking an aggressive view about what AI can and can't do. I believe AI is not yet ready to build enterprise SaaS-level software, but the jobs and tasks that humans consistently have to do on top of their SaaS software are perfect for AI to handle.

Greg Isenberg predicts "the most valuable skill in 2026 will be knowing when NOT to use AI." Human touch will trade at a 20x premium because the key is knowing which problems require human judgment versus automation.

Automatic Model Switching Is Here

GPT-5 introduced automatic model switching, which Grok already demonstrates. The system routes queries to different models based on complexity and requirements; simple questions go to lightweight models, while complex analysis triggers specialized systems. We're no longer choosing between one hammer and nothing.

Track your AI interactions today. Document problems you're solving with AI and those you're handling manually, and you'll discover patterns most organizations miss.

The Microsoft study Codie Sanchez references provides concrete data. Interpreters and translators face 98% AI applicability scores while phlebotomists show 6%, proving that specific scoring beats binary thinking.

No job is entirely secure, but every job can be augmented and scaled.

Your AI Evaluation Framework

Audit your current processes immediately. Map every significant business process and score on three dimensions: repetition frequency, decision complexity, and human relationship requirements.

High repetition, combined with low complexity and relationships, makes an immediate AI candidate, while nuanced human judgment or relationship building remains human-led for now. We covered this a few weeks back here:

Focus on the tasks layered on top of your existing systems, the reports you manually create from your CRM, the data reconciliation between platforms, and the meeting summaries you write after every call. These repetitive tasks on top of SaaS are AI's sweet spot today.

Require demonstration in hiring. Every candidate should demonstrate how they'd use AI to address a relevant problem, because understanding appropriate application separates strategic thinkers from binary followers. Give candidates a real challenge from your business and watch how they determine whether AI helps or hinders the solution.

Build tiered routing systems now by creating decision trees that match problem complexity to model capability. "What are your store hours?" doesn't require the same horsepower as strategic planning questions.

GPT-4 costs $0.27 per 1,000 tokens while Claude Haiku costs $0.0048, meaning routing simple requests to the correct model cuts costs by 98%. I’m not saying you have to build a router, but you should have a basic understanding of personal and work model routers, knowing which models to use for specific purposes.

It's like sending a postcard versus overnighting FedEx; both deliver messages, but one costs pennies while the other costs dollars. Use lightweight options for routine tasks and reserve premium service for what matters.

Test AI agents in controlled environments. AI agents hiring other agents sounds absurd until you see it happening in narrow domains, where marketing automation platforms trigger design tools based on performance metrics. Start with low-risk automated decisions and scale based on results rather than speculation.

Document the 20% where AI fails. Track every instance where AI tools produce inadequate results, because these failures map capability boundaries and identify where human expertise commands premium value. Your competitive advantage lies in these failure zones.

The Premium on Human Judgment and Connection

A real opportunity exists in selective application. Virtual companions will become a $50B market, as Isenberg predicts, because AI serves specific emotional needs at scale.

Human judgment and connection grow more valuable as determining when AI helps versus hurts becomes the differentiator. The person who knows when to deploy AI versus human expertise commands the highest premium.

Your hiring should reflect this reality by evaluating whether candidates know when to use AI and when not to. Candidates should demonstrate against real problems how, which, and why they would use AI.

The path forward requires understanding that AI is duct tape that works for thousands of applications. The skill is knowing which ones.

At Yext, we're using internal teams to analyze data proactively. We're tracking AI performance and analyzing the data that influences AI behavior across search, discovery, and trust signals. Check out the first Journal, published last week.

Companies that win by the end of 2025 will choose carefully where they apply their AI capabilities, recognizing that it excels at augmenting existing systems but is not yet ready to replace them.

We don't need to protect human creativity because AI augments creative ability. What we need to protect is in-person human interaction.

Matthew Lieberman's research shows our need for social connection equals our need for food or shelter. "Being socially connected is our brain's lifelong passion. It's been baked into our operating system for tens of millions of years” (Social: Why Our Brains Are Wired to Connect, 2013). Human connectedness drives everything because our brains evolved for social interaction, and we conquered every other species through it.

AI presents three paths for human connectedness:

1. Interrupt human connection

2. Replace it

3. Create more time for it

Using AI for routine tasks frees us for what matters, more time for actual human connection and the social bonds that define us.

I'm betting on the third path.

Start building that judgment now. The best builders know when to use duct tape and when to reach for something else.