I use artificial intelligence (AI) in two very different places. At work, it helps me build data products at a global bank, inside a system of specs, reviews, and quality gates. At home, it runs this website, helps me write books, and drafted this very post from an interview with me. Same underlying technology. Completely different relationships. And the gap between the two has taught me more than either side alone.
The Short Leash at Work
At work, AI operates on a short leash, and it should. The data my team ships feeds real decisions with real money behind them. A wrong number is not a typo, it is a loss. So the discipline is heavy by design: every requirement in a spec traces to evidence, agent output gets reviewed the way you would review code from a talented new hire, and machine-verifiable gates check quality before anything moves. The agent does not get to guess, because the spec is written so it never has to.
From the inside, this can feel slow. It is not. It is the speed limit that lets you drive at all. The short leash is not distrust of the tools, it is respect for the blast radius.
The Long Leash at Home
At home, the leash is long. This website mostly runs itself: an agent surveys my repositories to keep the work sections current, publishes a post most Sundays, and fixes its own plumbing when something breaks. When it fails, the worst case is a typo in public or a missed Sunday. Failure is cheap, so I let agents attempt things I would never allow at work: restructure the whole site, redesign the pipeline, interview me and draft in my voice.
That freedom is not a luxury, it is the point. Home is where I learn the actual ceiling of these tools, because the only way to find a ceiling is to let something run until it hits one. Every capability I now trust at work was first proven on my own time, where the stakes were mine alone.
Each Side Fixes the Other
Here is what the home lab teaches the office: the tools are further along than your org chart believes. If the locked-down work version is your only exposure, you will quietly underestimate what agents can do and keep humans on work that no longer needs them. My instincts about what to delegate did not come from a vendor demo. They came from cheap experiments on my own projects, run on nights and weekends, where being wrong cost nothing.
And here is what the office teaches the home lab: rigor travels. My personal projects got dramatically better when I brought the work discipline back with me. This website has a real test suite, evidence standards for its content, and a human review on every merge, because I learned at work that agents amplify whatever discipline or sloppiness you feed them.
So my advice is to own both ends of the leash. If your employer's AI is the only AI you touch, give a model a real job in your personal life this weekend and watch what it can actually do. If your personal AI is all play, borrow one habit from the office: write down what done looks like before you start. The people who will do this next decade well are not the ones with the best tools. They are the ones who learned, in both settings, exactly how much slack the leash should have.