I run a solo agency with four specialists on one isolated cloud computer. Not a chatbot. Not a Zapier chain. A Research agent, an Analyst, a Content specialist, and a Social agent — each with duties on a clock. Below is a real Tuesday. Names changed. Numbers are not.
6:04am — production starts before you do
Four duties. Zero laptop opens.
At 6:04am the Research specialist opens 17 competitor pricing URLs through a residential VPN — only the pricing paths, not whole domains, because VPN minutes are capped per plan on Growth and Pro. It compares against last Tuesday's CSV at ~/files/reports/competitor-prices.csv, screenshots one diff, and writes a two-line summary. At 6:09am a second duty fires: lead research for an 11:30am discovery call. The agent reads the prospect's site, LinkedIn, and two review pages. Output lands in ~/files/research/acme-brief.md.
A solo founder in Bristol ran the same stack last month. She used to wake at 6:45 and open eight tabs before breakfast — competitor check, call prep, inbox scan. Ninety-three minutes, every weekday. Then she missed a rival's Sunday price drop because weekends were not in her manual routine. She found out Monday afternoon on a renewal call. Now her phone buzzes at 6:52 with "New low: £89/mo — screenshot saved." She reads it in bed. Done.
Insider detail: Route anomaly-only WhatsApp alerts for the competitor duty. If you notify on every run, you train yourself to ignore the phone. I only get a message when a price field changes or a duty fails — not when 16 URLs stayed flat.
9:18am — the client call you actually prepared for
One call. Forty minutes of judgment.
My only fixed meeting today is 9:30am with a B2B SaaS prospect. At 9:18 I open the PWA from my phone — bottom dock, Files app, skim the brief the Research agent wrote at 6:09. Their competitor launched a case-study hub last Thursday. Monitoring flagged it in Tuesday's 6:04 run. I spend eleven minutes adding two questions the agent could not know: budget range from a mutual contact, and whether their CMO cares more about pipeline or brand.
Before this stack, I walked into calls with a half-filled Notion page and guilt. A three-person consultancy in Austin told me the same story: 45 minutes of pre-call Googling, repeated for every discovery slot, four times a week. Twelve hours monthly on research that followed the same checklist every time. Their Calendar agent now triggers Research 90 minutes before each call. Founder time dropped to 12 minutes of review. Close rate went up — not because the agent is charming, but because nobody asks "remind me what you do again?" on minute three.
The Analyst specialist does not run on Tuesday mornings. Friday at 4:07pm is its slot — client report duty. Tuesday is for selling, not formatting PDFs. That separation is deliberate. If you want the full Friday recipe, see our weekend setup that automated Monday morning admin.
Solo agency founders treat agent schedules like one long workday. Wrong frame. Your duties should mirror client rhythm — competitor intel before standup, content while you are in meetings, social scans after publishing windows close. Stack four duties at 6am and you will ignore half of them by week three. Stagger by when the output is actually useful.
2:11pm — content drafts waiting in the file workspace
Lunch ends. Drafts do not wait.
The Content specialist ran at 2:00pm — duty wired to read ~/files/reports/competitor-prices.csv and this week's monitoring notes. It drafts one LinkedIn post and one client newsletter section referencing the £89 competitor move. Files save to ~/files/content/. I open the DOCX inline editor on my phone, fix one headline, save. Same path. The Social agent reads it at 5:00pm if I approve.
A five-person shop in Toronto tried the opposite: Content agent scheduled for 6:30am with everything else. By 8am they had four notifications and zero reviews. Production piled up while they were still in inbox triage. Moving Content to 2:00pm — after the morning call block — cut ignored drafts from six per week to zero.
Cloudia built this specialist in one conversation: "Draft from competitor diffs every weekday at 2pm, save to content folder, never publish without my approval." She enabled the file skills, pointed the duty at existing paths, and left publishing disconnected until I connected LinkedIn in the Launchpad. Total setup: 68 minutes on a Sunday. For how specialists fit together inside the OS, read hiring specialist AI agents.
5:22pm — social monitoring before you close the tab
Last duty. Five minutes of human review.
At 5:00pm the Social agent checks connected X and LinkedIn accounts — client pages and three competitors. It logs engagement deltas to ~/files/social/weekly-metrics.csv and queues approved posts from the 2:00pm draft. I get a WhatsApp only if a competitor post outperforms ours by more than 40% on the same topic. Tuesday's message: a rival's carousel hit 12k impressions. Worth a rewrite. Not an emergency.
I used to end Tuesdays at 7:30pm manually pulling screenshots into a deck for one retainer client. Eighty-five minutes. Same data. Every week. The Social duty runs in 9–14 minutes on the isolated cloud computer; I spend five reviewing the CSV on my phone before dinner. The client still pays $650/month for the summary. The production cost is a Growth plan at $19/month plus review time — the same economics agencies use when they bill the same rate and agents do the work in minutes.
Insider detail: The visible browser matters here. When a competitor changes their LinkedIn layout, the agent adapts — it is not a brittle API field mapping. I watched it happen last month: new carousel format, agent still scraped the caption and image count. A Zap would have broken. This one did not.
What I still do myself — and why that is the point
Reframe: the goal is not zero founder hours. It is zero founder hours on checklist work.
I still write strategy decks. I still negotiate scope. I still fire clients who drain margin. I still approve every outward-facing post — the agents draft, I decide. Tuesday's total agent-driven production: roughly 4.2 hours of work I used to do by hand. My logged founder time on the same deliverables: 38 minutes of review and one 40-minute call.
Remember the Portland founder at 7:03am? By 5:30pm she has taken one discovery call, edited two headlines on her phone, approved one LinkedIn post, and closed the PWA without opening a laptop. Her agents ran four duties on schedule. She ran one — the conversation only she could have. That is the solo agency stack: not replacing the founder, but removing the production layer that pretended to need her at 6:45am every morning.
If you are building this from scratch, start with AI agents for marketing agencies and stagger your first two duties using our guide to scheduling AI agents 24/7. Launch the stack at app.cloudaxis.ai — free to start, no credit card.
Your agents have a schedule. You should have one too — and they should not be the same calendar.
Frequently asked questions
How many AI agents does a solo agency actually need?
Four covers most solo retainer work: Research for intel and call prep, Analyst for weekly client reports, Content for drafts, Social for monitoring and publishing. You can start with two — Research at 6am and Content at 2pm — and add Analyst and Social when the first duties run cleanly for four weeks. A solo founder billing $8k–$15k/month typically breaks even on Growth ($19/month) in the first week if manual research alone ate 6+ hours.
What does an AI agent daily routine look like on a phone?
Install CloudAxis to your home screen (PWA). Morning: read the 6:52am WhatsApp summary. Midday: open Files, skim call briefs and content drafts. Afternoon: approve posts from the Social queue. You do not need a laptop for any of that — the file editors are touch-optimised. Push notifications work on iOS 16.4+ if you prefer lock-screen alerts over WhatsApp.
Can one person manage AI agent duties for multiple clients?
Yes — separate file paths per client (~/files/clients/acme/, ~/files/clients/nova/) and one duty per workflow, not per client crammed into a single prompt. The Research agent can run sequential client checks in one duty if the CSV lists all URLs with a client_id column. Most solo agencies cap at 4–6 retainer clients before hiring a human account manager; the agent stack scales to that range on Pro ($39/month) without adding headcount.
AI agents for marketing agencies · Weekend setup for Monday morning admin · Scheduling AI agents 24/7