So Now the Company Wants You to Use AI
"We should be exploring AI," someone said. What does that mean? For a lot of us, it means we’re now unofficial AI consultants, whether we asked for it or not.
The Corporate Predicament
Executives want innovation using AI, but often don’t know what that means. There’s a push to adopt tools or “do something” with AI, not because it solves a clear problem, but because everyone else is doing it. AI is becoming checkbox tech:
- ✅ “Are we using it?”
- 🤷 “Is it helping us?”
The result is a gap: there are meaningful ways to leverage AI in real workflows, but policy friction and lack of clear goals leave teams stuck between skepticism and hype. We can't just turn on copilot PR reviews and call it a day.
The reality for software developers
We're living in the wild west. Some people are cruising through cyberspace while others haven't even left town yet. StackOverflow is an old paradigm that is on life support. AI won't default to telling you that you're an idiot and closing your question, but that means there's less overall questions asked in the human-to-human open online spaces.
Ziggit, which is the main forum I use for Zig, has new rules around AI. These rules are reasonable community norms. Nobody wants to read your AI generated code that doesn't compile. Expectations for AI use needs to be set up front in human-first spaces. I know that sounds right out of Blade Runner, but that's where we are at today. We need to have policies that keep our communities healthy.
New skills to acquire
I highly value the master-apprenticeship model. Having a mentor who helps you maneuver towards your lofty goals? Amazing. Being that mentor for someone else? Incredibly rewarding. Both of these dynamics are relevant when working with AI.
I usually treat Cursor or Copilot as my apprentice. I find myself mentoring the AI by defining how I want it to work through patience and guidance. In another window, I’m the apprentice asking for help understanding a new library or stack.
If you’re not using something like .cursor_rules.md, you’re missing out. It’s like discovering linters for the first time. Clear expectations, style, naming—codified and enforced without micromanaging. It’s a great place to reinforce that totally often thought about team working agreement or style guide, instead of letting them fade into oblivion.
This idea—embedding context and behavioral constraints for tools—is also driving efforts like the Model Context Protocol (MCP). Tools like Honeycomb and ClickHouse are building support for MCP so people can have open-ended conversations about production. It’s a powerful idea, and yet another tool in the AI toolshed.
I foresee this frontier here being very hard to keep up with. New skills for improving or applying AI will keep surfacing–new tools, workflows, and jobs. It’s moving too fast to give solid advice beyond: hold on and try to keep up.
Multi-Classing and the AI Edge
People with expertise across domains—what I call multi-classers—already have an edge. Now, pair that with an AI agentic workflow and things get wild. You’ve got someone who knows how to code, manage projects, write docs, and wrangle models? That person can spin up small, powerful systems fast. But before you go full puppet master and wire up a fleet of agents doing everything autonomously, a warning: it still takes a ton of skill and practical knowledge to make that dream work.
You might be better off finding collaborators who complement your strengths before you try to solo the AI-dungeon with a rogue team of LLM-powered minions.
Final Thoughts
It's all together exciting and frustrating to have yet another layer of ideas and tools to learn. I think sticking to the fundamentals of good judgement, clarity in purpose, and being excellent to one another (human or not). So yeah, now your company wants you to use AI. It's more than checking the box or using ChatGPT as a Google replacement. Do you understand your options? Are you paying attention to the console cowboys in cyberspace?