Your store can open in week 2. Agents run it from the first order.

You want your Shopify store run by the AI company in your document, and you want to sell fast. With one or two products, the store does not need six weeks. It can open at the end of week 2, once your product and photos are ready, and agents run it from the first order. Something new goes live every week after that. By week 6, Customer Experience, CRM, and Marketing run on agents with real orders behind them. The rest of your chart can follow by the end of January. Where a role depends on your traffic, your creatives, or a paid plan, the chart says so.

Your Shopify, run by agents

Shopify is item 1 on your list: store, orders, products, customers, inventory. It is also where every other agent gets its facts. The store agent connects in week 1. I set up the store with you in weeks 1 and 2, and it opens as soon as your product and photos are ready. Here is what the agent does in each of those five areas.

A live win every week

Nothing waits for a launch day. Each week ends with something working in your store or your Slack, and each week is its own Upwork milestone. If kickoff is Monday 5 October, the store opens the week of 12 October and ads start the week of 26 October.

Six fixed-price milestones on Upwork, total $18,000. Each is funded when it starts and released when it is live and you have seen it work. If a week slips because of me, it is not released until it is live. If it slips because an account or an answer is late on your side, the date moves and the price does not. The week 2 opening depends on your side: product in hand or on pre-order, product photos, Shopify Payments approved, and the domain. If your stock lands later, week 2 is released when the store is ready with the password on, and the password comes off the day stock lands. The weeks that need real orders move with it. Everything else keeps its week. Orders before week 4 come from your list and your own network, so they are few, and your inbox covers the handful before support goes live in week 3. Google approval of the Shopping feed is Google's call. If Google holds a new store for review, the agent explains why and drafts the fix the same day. Assumes your Shopify store, Gorgias, Klaviyo, ad accounts, and Slack exist by the end of week 1, Triple Whale, Littledata, and KnoCommerce by week 2, and someone on your side answers questions the same day.

What depends on your product and your creatives

AI is strong at some parts of this job and weak at others. I would rather say that now than find out together in week 4. Here is the line.

Claims. For a wrinkle product the wording matters. "Reduces the look of fine lines" is a cosmetic claim. "Removes wrinkles" or "builds collagen" makes it a drug claim in the FDA's eyes, and a device has its own rules. The claims check keeps copy on the right side, but it can only use what you can back up. In week 1 I ask what backs each claim, and a regulatory reviewer on your side should sign off on the main ones.

By week 6: the three departments you asked for

The point is fewer things on your desk. Anything reversible or read-only runs on its own from the day it goes live. Approvals stay only where money moves or a customer is upset. Each approval is one tap in Slack, and every tap trains the rule that removes the next one.

Measured from the first visitor

Some data only exists if the tool is on when it happens. A pixel counts only the visits after it is installed. A survey asks only the orders placed after it is live. So the tracking goes in the week the store opens, and your first sales and Black Friday get measured.

Jarvis sits above all three in Slack. It posts one morning brief grouped by department, routes your requests, and escalates anything an agent is unsure about. A watchdog checks every agent ran and every integration answered, and tells you if not.

How you will know it works

The command center tracks these from the day each piece goes live, per brand: share of tickets closed without a human, first-reply time, approvals per day and how often you edit a draft, Klaviyo flow revenue, ROAS by channel, and model cost per order. If an agent is not earning its place, the numbers show it within weeks.

After week 6: the whole chart by end of January

You start selling in October. The fastest way to the rest of your chart is three short builds, one Upwork milestone each, picked one at a time. On the Operator plan the same items arrive about one a week instead. Black Friday comes first. By then you have about six weeks of sales and four weeks of ads behind you. Anything read-only or easy to connect is already in the launch build. What waits needs traffic, sales history, or write access to your store.

Dates assume kickoff on 5 October 2026. The store date moves with your stock and photos. No build work from 21 December to 1 January. The agents keep running and I keep watching them. From February, the Operator plan keeps it improving: new tests, new agents, and fewer approvals as each agent earns it. Brand 2 can start in February. You can reorder anything between builds.

What you asked for, what you get

Your org chart, month by month

This is your chart from your document, every role in it. In the build, each specialist is a skill inside its department agent, so it needs no separate server. You still see every role: each action in the log is tagged with the role that did it. Same reporting lines, far fewer moving parts, lower running cost. Live means the role does its job on a schedule or when asked, and you can see its work in the log. A small letter on a role means it depends on something outside the build. The key under the chart says what.

• ^a Needs traffic A test needs enough orders to call a winner. On a new store one test can take weeks, so early tests are big changes, like the offer or the top of the product page. Small tweaks wait for traffic.
• ^b Needs customers first Built and tested by week 6. It has work to do once there are tickets, reviews, refund requests, or buyers past their reorder date. Until then it has little to say.
• ^c Only with subscriptions Only if you sell subscriptions through Recharge or Skio.
• ^d Early read A few months of sales gives a first estimate. Treat it as a guide until there is a year of data, and plan cash on something firmer.
• ^e Needs Shopify Plus Most checkout changes need Shopify Plus. On other plans the agent works on the product page, the cart, and the thank-you page.
• ^f Needs your creatives The agent writes the text, the briefs, and the schedule, and reads the results. The images and video come from you or your designer.
• ^g Drafts, you decide The agent drafts and researches. Strategy, supplier messages, and anything that moves money wait for you.
• ^h Quick, with approvals early on Chat replies follow the same approval rules, so early on some wait for your tap. For instant chat answers, Gorgias sells its own AI agent.
• ⁱ Needs your accounting Needs read access to your accounting or bank data. That is not on your tool list yet.
• ^j Needs a paid API plan The tool's API is a separate, higher plan. Details under the tools list, so you can decide before you buy.

All 41 tools, by month

What actually runs on your server

Not a chat window and not my laptop. One VPS in your name, Ubuntu and Docker. Each agent is its own container running Hermes Agent, an open-source (MIT) agent runtime from Nous Research, under a process supervisor. It runs 24 hours a day and restarts itself in seconds if it crashes. Each container has its Slack connection, a scheduler for its timed jobs, and its own tool servers (MCP) for the platforms it works with.

This is the setup I run for my own ecommerce brand today: five always-on Hermes agent containers on one VPS, an operations agent with an approval-gated executor on Shopify and the 3PL, an email agent, and a shared Postgres + pgvector brain. Nine scheduled production agents, 62 automation scripts, four sales channels.

That is the part that runs. The build side has its own team: while I work, several Claude Code and Codex agents run at once, and they hand research, code, and tests to sub-agents in parallel. That is how one person builds this in six weeks.

A-Z architecture

Top to bottom. You talk to Jarvis. Jarvis routes to a department. Department agents propose actions. The gate decides whether an action runs now or waits for you. Approved actions go to your tools. Everything is written to the brain, which every agent reads before its next task.

Your learning loop, by week 6

The example from your document, step by step, with what runs by week 6.

What runs on its own, the rules

Every action an agent can take is classified once. The classification lives in the brain, so you change it without a code change.

Price

Start with the launch build

Each week is its own fixed-price Upwork milestone, released when you have seen it work. Watching and fixing the live agents is included while the build runs. If you stop after week 6, you keep everything built.

After week 6: keep it running and growing

Faster option: the whole chart by the end of January

Same weekly rate as the launch build, $3,000 a week. You pick them one at a time after week 6, and each is its own fixed-price Upwork milestone. Watching and fixing the live agents is included while they run, so the Operator plan starts in February.

Your other brands

The platform is multi-brand from week 1: one VPS, one brain with a brand tag on every record, one command center with a brand switcher. Rolling out a brand means connecting its accounts, loading its brand book and product facts, running the agents supervised for a week, then letting the proven ones run. Budget $6,000 per additional brand, two to three weeks each, as its own Upwork milestone. Best start is February, once your first brand has the whole chart. A brand that already runs Klaviyo and Gorgias takes the longer end, because we map what exists first.

Running costs, paid by you to the providers

Honest notes

One person, start to finish. You get me on the weekly call and in Slack, with no account manager in between. You own the code, the server, and the data. Cancel the plan and everything keeps running. Anyone who can restart a container and read a log can take it over.

Next step