It was a transition, but I wouldn’t say it was that difficult. It helped that coming back meant working from home. Three more weeks and then it’s back to regular programming. Just in time for the kids to be back in school.
I do wish it were longer. I feel I have a bit more focus at home, but not for the same reasons as before.
In the past, pre-AI, working from home meant being able to sit at my desk undisturbed, with my ultrawide screen so I could have all my code up where I wanted it, my ergonomic split keyboard properly tilted and configured for utmost comfort, the occasional walking pad under my desk to keep me moving throughout the day.
My main gripe with working at the office was blocking out the noise from the open space layout we have.
Recently, the reason I can’t really focus on a task is the fear of being seen as idle. Whenever I set an agent off to work, I don’t really have much else to do, so either I start off another agent on a parallel task or I go snoop around the logs to find something else to fix. Pretty soon I have 7 or 8 active threads happening, and sometimes I forget some of them.
At the end of the day, I’d only have moved the needle a little bit on each of them.
When I work from home, I fire off the main agent, and while it’s working, I think twice about starting up a new one. Instead I go make myself coffee, trigger a load of laundry, get a glass of water, do some micro exercises. There are all these things I can do in the 3-5 minutes it takes for my agent to come back for the next step. During those moments when I’m up, my work mind is still on the same task, thinking about other approaches, planning in my head the next steps for when the agent comes back.
I still fire off 2 or 3 other agents on quicker tasks, but I have less temptation to fire off something that I know will take some time to resolve.
The definition of working has changed in the agentic era. A lot of the work is actually thinking and not typing.
You take a step back and focus on the end goal and the bigger picture. You’re not in the weeds typing out code, remembering syntax, or googling keywords to help you out.
Instead, you start thinking about how you can ensure that your agent’s output is correct.
If you think you’ll be able to review the files, sure, you can do that. But in my experience, things at that level will look syntactically correct. Whether the hundreds of syntactically correct files actually result in the right behavior, well, that’s a context overflow for the human mind. There’s no way to do that ‘manually.’
Instead, you start thinking about what guardrails to put up. What’s an integration test you can set up to verify the code generated by your AI does what it’s supposed to do?
If that’s too difficult, you fire off another agent to figure out how to make the system a bit more testable.
This is what a typical day of work looks like for me now, and a lot of it is thinking, playing out scenarios in my head. Have we thought of this case or that case? Then I build the guardrail test for it (with an agent, naturally).
In a way, that’s better.
If you have the luxury of starting a project from scratch, however, I think that workflow would look very different. If you can plan things ahead with event modeling, for example, it’s just a matter of getting the agent to implement against the event model. If you’re lucky enough to set up a clear event model from the get-go, you can easily design tests that verify it, and verify that the agent produces code according to spec.
Either way, a huge bulk of the time no longer lies in the coding, but in the thinking and planning. Whether you’re planning with event modeling, planning in your mind, or discussing with a planning agent.
I’m thinking about how I’m going to go back to the office when it’s time. I don’t think starting several threads is a great look.
I need to figure out what other micro breaks I can take. Maybe it’s developing a coffee addiction to force myself to get up and wait for the coffee machine, and use that wait to plan the next tasks for my maximum of 3 active agents.
being in tech 