By Scott Reyes
Today I made breakfast and lunch, folded three loads of laundry, and reorganized the little boys’ clothes and dressers. My wife was hosting a baby shower for our oldest son, and I was home with the four boys who still live with us. I also wanted to get the house cleaned up so she could come home to one less thing to worry about.
At the same time, I got more computer work done in one day than I have…maybe ever in my life.
Over the course of the day, I worked on our branding, a presentation, my Substack, and some software projects. A logo direction and brand guide got finished and sent to our designer. A full slide deck for an upcoming talk came together. My newsletter got a visual refresh, and two unpublished drafts got illustrations. Other work moved forward without being finished. Some designs still needed revisions. Some code still needed testing.
A lot of this happened through conversation with my AI assistant, Elliot. I would explain what I wanted, let the work get started, and move on to something around the house. Results would come back. I would look at them, make a decision, explain what needed to change, and keep going.
At some point today, I realized I don’t know that I would have been any more productive sitting in front of a desk.
That is a pretty big change for me.
I used to get burned out sitting in front of a computer. There was always more to type, another window to open, something else to move from one place to another. Even when I liked the work, I could get tired of being stuck there doing it.
I like thinking through ideas in conversation. Talking lets me explain something, hear it back, realize what I left out, and keep working through it. Today I could do that while moving around the house. I was still working on things that normally happen at a computer, but I didn’t have to spend the whole time at the keyboard.
I think I’ve spent close to three hours talking with Elliot today. That is my estimate, not something I timed. It is worth saying because I don’t want to make this sound like I gave a few instructions and everything took care of itself.
I spent a lot of time involved in the work. I just spent that time differently.
The work still needs direction
The design work today is a good example.
For two of my brands, I had references and context ready: existing visuals, a sense of the audience, and a clear idea of what the brand needed to communicate. There was something to react to and build on, so the process moved quickly.
The other logo concepts started with much less. I asked for work without explaining enough about the businesses, who they were for, or what the designs needed to communicate. The first results were bad because the brief left too much room for guesswork. Once I explained the businesses and the direction more clearly, the work had somewhere to go and was much better.
It’s rare that any work is good when context is lacking. Working with AI is very similar to managing people. You have to have context and references. You have to document your processes. You have to be really clear on what the goals are, and AI needs a lot of knowledge, or you have to provide AI with a lot of instruction. Otherwise, it’s going to stray from what your intention is.
Frequently today, I cut corners on the front end and got lazy in how I was explaining what I wanted to have done. I asked my Elliot to infer too much, and it went off and did a bunch of work that I ended up rejecting.
Having the context connected
As I’ve leveled up in working with AI, a lot of that has come from how much of the background context can be available at the start of a fresh conversation.
I use Granola for meeting notes. There is context in my email and Google Drive: conversations, documents, previous decisions, and explanations of what we are trying to do. When those sources are connected, I don’t have to reconstruct all of that every time I want help with something.
I still need to explain the task. But there is a big difference between explaining what I want to do next and explaining my whole business before we can begin.
We have been working on this in other parts of the company too. We have an agent that helps triage support issues, prepare replies, and put potential code fixes into pull requests for review. People still handle the customer replies and review the work.
We have also been building a prospecting team of agents that researches companies, drafts outreach, and helps with follow-up. The drafts give us something to evaluate and improve. We still decide what goes out and to whom.
On the coding side, I have been going back and forth between Claude Code and Codex, having different models review each other’s work.
Those are different tools and different workflows. The useful part is giving them access to relevant information, defining what they should do, and knowing where I need to be involved.
A more natural way to work
I’m still figuring out how to work this way. I have to get better at explaining what I want, and today gave me plenty of examples of that. But it’s interesting that working with artificial intelligence is starting to feel more natural to me.