The Real AI Skills Gap Isn't What You Think

I posted something like this on LinkedIn a few weeks ago, but wanted to revisit it in a longer form…

For every engineer on my team doing advanced work with AI (building MCP servers, creating custom agents like Ralph and GasTown, automating complex workflows) there are ten who are overwhelmed by the noise. They’re not sure what tools they should be using. They don’t know why Sonnet 4.6 is better (or worse) than Gemini 3.5 Pro. They’re still using Copilot as a glorified autocomplete in their IDE.

And that’s a leadership problem, not a skills problem.

The AI landscape is full of noise. If you listen to the Sam Altmans and Dario Amodeis of the world, you’d think we’re already being replaced by sentient robots. On the other hand, you have people convinced the AI bubble will burst and this all goes away. The truth is somewhere in the middle. The bubble will absolutely burst (the OpenAI → NVIDIA → Anthropic circle looks like a Ponzi scheme right now). But the underlying technology isn’t going anywhere. Open source models like Kimi K2.5 prove that. These tools will remain in some form, and engineers need to accept that and learn to work with them.

As engineering leaders, our job is to create clarity in this chaos. Sometimes that means stepping back from feature delivery and investing in foundational knowledge. Making sure our engineers have room to explore new tools and aren’t drowning in hype.

Because here’s the thing: our job as engineers has never been “writing code.” It hasn’t changed since they built the pyramids. We solve problems using the tools we have at hand, weighing reward against risk, and making technology work for us.