I ran a one-person practice on three hours of sleep for most of a year. It produced real output and it almost broke me. The AI studio morning routine below is what I rebuilt after I stopped pretending three hours was a plan. It runs the agent council, the day's first build loop, an inbox sweep, and a walk before 10am. No 3am starts. The routine survived a client escalation and a cross-country trip, which is the only test that matters.
These are field notes. Four dated entries from a recent stretch. What I tried, what held, what I changed.
Agent council triages the inbox in six minutes. The rest is one real build task.
2026-04-15: the 6am start that replaced the 3am start
Got up at 5:55am. Coffee on before looking at anything with a screen. The first 40 minutes of the day are deliberately unplugged. Journal for five minutes. Read something non-work for twenty. Shower. The rule is that no agent, no Slack, no client channel touches my eyes before coffee is done.
At 6:40am the first work block opens. I walk into the studio and fire the agent council. It has a standing prompt: summarize anything that came in overnight across the client channels, flag anything the CEO persona would want me to see first, draft replies for anything routine. Six minutes later I have a morning triage doc with three buckets: send today, decide today, ignore today. The council is doing the reading so I can do the deciding.
The block runs from 6:40 to 8:10. This is the hardest work of the day. The one real build task. Today it was reworking a Shopify section to pass a mobile Core Web Vitals target. Claude Code drove the typing. I drove the architecture. No meetings, no calls, no Slack open.
At 8:10 the first break. Twenty minutes. Eat real breakfast, not coffee with protein powder. No phone at the table.
2026-04-16: what the agent council replaced
Tried something different. Ran the routine without the agent council to see what broke.
What broke: the first hour of work became inbox. I spent 50 minutes reading client Slack threads, trying to figure out which ones needed me before the build block started. By the time I got to the real work at 7:40am I had already spent my best cognitive hour parsing other people's questions.
The agent council's job is not to replace me reading the messages. It's to compress the reading time to something that doesn't eat the peak cognitive window. Six minutes of agent work saves about 45 minutes of human work, and it saves the 45 minutes that's worth the most.
This is the clearest argument I have for agent leverage in a solo practice. Not that agents do the work. That they protect the window where I do the work.
2026-04-17: the walk, and why it's non-negotiable
From 8:30 to 9:10 every morning I walk. 40 minutes, outside, no headphones most days. This is where the day's plan gets finalized.
I used to think the walk was optional. A nice-to-have. I dropped it for a week in March and my afternoon output fell off a cliff. The walk is not for health, though that's a side effect. The walk is where my brain does the messy synthesis work that the morning build block started. I come back with the afternoon's build task already shaped in my head.
Today's walk produced the structural decision for a product page rework I was stuck on. Not because I was trying to think about it. Because I stopped trying.
I write about the broader solo-practice shape in the creative-tech operator playbook; the morning routine is one instance of what that role looks like day to day.
2026-04-18: the 9:30am block is for humans
9:30 to 10:30 is the second work block, and it runs differently from the first. First block is alone with the agents. Second block is human-facing: the two or three calls I take in a day, the async replies that need judgment, the decision frames that go to clients.
The agent council drafts the outbound communications but I rewrite most of them. Council drafts are a starting point. At this point in the morning the draft is either a useful skeleton or it's not, and I know within thirty seconds which one. I've written about persona calibration in the agent council pattern article; the test of whether a council is shipping-ready is whether its drafts save time at 9:30am on a Tuesday.
“The agent council's job is not to replace me reading the messages. It's to compress the reading time to something that doesn't eat the peak cognitive window.
”
What the week taught me
Three patterns hold across every iteration of this routine.
The morning build block is sacred and has to come first. If I move it to after lunch, the quality drops. Not a little. Measurably. The work I produce before 8:30 is materially better than the work I produce at any other time of day. Every version of the routine protects that window.
The agent council is a pre-processor, not a replacement. It runs against the inputs that would otherwise consume my peak hours. When I've tried to use agents for the peak hours themselves, the output gets worse, because the interesting work is the part where a human decides what matters.
The walk is a feature, not a break. Forty minutes of no-input time between the solo block and the human block is where integration happens. Skip it and the afternoon compounds the friction instead of building on the morning's work.
The tool layer
The agent council runs on the setup I documented in the 200 dollar a month AI tooling stack. Claude Code orchestrates the morning triage. The council is five named personas, each tuned to surface a different lens on overnight activity. The build block uses the same Claude Code environment, just pointed at the day's specific task. No separate tools for separate jobs. The cognitive cost of context-switching is the bottleneck, so the studio runs everything through one workflow.
The infrastructure itself is the stack I wrote up in the one-person studio stack. Vercel for the site, Supabase for data, Stripe for billing, Resend for email, Claude Code for the work. The morning routine runs on top of those without ceremony.
Frequently asked questions
Do I need agents to run a routine like this?
No. The shape of the routine (solo block first, walk, human block second, afternoon build) works without any agent tooling. The agents compress the inbox triage work that would otherwise eat the first hour. If you don't have an agent setup yet, the routine still works; you just spend more time on inputs and less on outputs.
What if my kids wake up at 6am?
The routine I described assumes the 5:55 to 10am window is yours. Many people do not have that. In a family-first schedule, the solo block moves to wherever the uninterrupted window actually exists, which is often 8pm to 11pm or 5am to 7am. The structure (solo first, input window later, movement in between) is more important than the clock time.
How did you get off the 3am schedule?
I stopped treating sleep as optional. The shift took about six weeks. The short version: the morning block has to be worth protecting, which meant reshaping my pricing so I didn't need evening hours to hit revenue. Once that math worked, the sleep followed. I've written about the economics side in the One-Person Studio OS, which is where the pricing and scoping framework lives.
What if the agent council drafts a bad morning brief?
It happens. About one morning in seven the council's triage misses something important or over-flags something trivial. The fix is to read the raw inputs myself that day and recalibrate the council prompt that evening. The bad brief is information; it tells me where the persona context needs updating. See the agent council pattern article for the calibration approach.
What's the afternoon look like?
The afternoon is the second build block (12pm to 3pm), another walk, then the admin tail: billing, scheduling, writing for the site, light reading. I end the workday at 5pm most days and read or cook after. The boundary at 5pm is the other non-negotiable. The routine's whole point is that a solo practice that outships agencies has to be sustainable for years, which means I treat the evening like a hard stop.
Sources and specifics
- Morning routine documented from field notes taken 2026-04-15 through 2026-04-18 in Pacific Time Zone.
- Agent council triage time observed to compress roughly 45 minutes of human inbox work into about 6 minutes of agent work per morning, measured across a two-week stretch.
- Solo build block produces measurably higher-quality output than any other time of day based on a personal audit of commits, drafts, and decisions across Q1 2026; the gap is noticeable, not marginal.
- The 5pm hard stop followed a Q4 2025 burnout episode that started on a 3am schedule; the routine described here is the rebuild after that, not the original pattern.
- Pricing framework that made the 5pm stop feasible is documented in the One-Person Studio OS; the technical stack it runs on is in the one-person studio stack post.