Your Company Limited AI? That's Actually Making You a Better Engineer
When my company capped AI usage and made us take training on tokenomics, I got mad for about five minutes. Then I realized the constraint was forcing me to think like an engineer again.
Last week my company capped our access to premium AI models, limited GitHub Copilot usage in VS Code, and required everyone to sit through tokenomics training. My first thought: annoying. My second thought: this is exactly what we needed.
Here's what changed and why I'm actually grateful for the constraint.
The constraint forces real orchestration, not lazy prompting
Before the cap, I threw everything into one chat and hoped something happened. Research, planning, implementation—all in the same conversation. Wasteful. Inefficient. Lazy.
Now I have to think. The training covered context windows, input vs. output tokens, and when splitting into separate chats actually saves money and time. That's not boring compliance stuff—that's learning how to work with AI instead of just using it.
The difference feels significant. Unlimited chats made me sloppy. Constraints made me intentional.
Separate chats for separate phases
This was the concrete tactic that stuck with me:
- Research chat: explore the problem space
- Planning chat: design the solution
- Implementation chat: write the actual code
Each one starts fresh. You reference markdown files from the previous phase instead of bloating one conversation. When a chat gets too large, you generate a markdown summary and open a new chat with that file as context.
This isn't busywork. It's teaching you to structure your thinking the way good engineers always had to—before AI let us get sloppy.
The real skill is context management, not prompt writing
I said in my last video that the real skill is orchestrating, not typing. Now I see what that actually means: managing context, defining references, understanding how pieces connect, knowing when to invoke an agent and why.
That's not about writing better prompts. That's about resource management. Cost. Efficiency. The same skills you'd need running any system at scale.
The constraint forced me to develop those skills instead of just mashing the "ask AI" button.
What's still missing in enterprise
I wish my company would adopt terminal-based agents like Claude Code—where you can orchestrate agents from the command line, not just from VS Code or a browser. That's where the real cutting edge is. Enterprise AI and frontier AI have a visible gap right now, and I feel it every day.
But I get why companies move slower: cost control, security, stability. Those aren't bad reasons.
The takeaway
If your company just limited your AI usage, you might feel frustrated. Don't. Feel grateful. The constraint is teaching you to think like an engineer again—to manage resources, structure information, and orchestrate tools with intention instead of convenience.
That skill transfers everywhere. And it's something no amount of unlimited API calls would have forced you to learn.