AI Side Hustle: Stop Selling Hours, Start Selling Plans

Written with AI assistance. Figures without a traceable source were cut before publishing.

The same piece is posted as a thread on GitHub — that copy has a reply box under it, and this one does not.

The most honest numbers I have seen attached to an AI side hustle are small enough that nobody would put them in a headline: one operator, Xiaomin, clears about 3,873 yuan a month generating images, another named Aqiang runs 5,000 to 6,000 yuan a month, and the avatars they sell go for 9.9 to 39 yuan each. Do the division and you get a picture that the “quit your job” genre never shows you, which is a lot of small transactions handled quickly. I think those figures are more useful than any success story precisely because they are unimpressive, and because the path from there to something better is visible in the arithmetic itself.

The Entry Barrier Really Did Collapse

Start with what changed, because it is the one part of the hype that holds up.

The current generation of tools genuinely runs on copy and paste. For image generation, video editing and voiceover alike, the working method for someone with no technical background is to take a prompt someone else wrote, paste it, and adjust it until the output is close enough to ship. Playbook posts now ship with the prompts, the client-facing scripts for taking an order, and a pricing template, so the thing a beginner used to spend three weeks failing at is handed over as a file. No tools to buy, no framework to learn.

That is real, and it explains the 9.9 yuan avatar.

It also explains why that price stays at 9.9 yuan. When entry costs nothing, the output is a commodity, and the commodity price is set by whoever is most desperate. Xiaomin’s 3,873 yuan is not a low number because she works slowly; it is a low number because she is selling units of production in a market where units of production are free to produce. Every hour she stops working, income stops.

The Move That Changes the Math

The advice worth taking from these operators is a single sentence: stop selling time, start selling plans.

A plan sold to 100 people costs essentially what a plan sold to 1 person costs. That is the whole argument, and I’d argue it is the only structural difference between a side hustle that plateaus at 5,000 yuan a month and one that keeps going. Productizing a skill means turning what you do into an SOP template, a toolkit, or a course, meaning an artifact that leaves your hands and keeps earning without your calendar being involved. The revenue stops tracking hours worked.

Which artifact you make matters less than the fact that it is an artifact.

The version of this I find most immediately doable is the SOP. You already have a process; you refined it across a few hundred small orders; the refinement is the valuable part and it is currently stored in your head. Writing it down as a document someone else can follow converts a habit into inventory. A toolkit is the same conversion one step further along, and a course is the same conversion with teaching overhead attached.

There is a real objection here, and it is not the one people raise. The objection is not “who would pay for a document.” It is that most operators have not actually stabilized a process worth documenting; they have been improvising per order and calling it a workflow. If you cannot write the steps down, that is the finding, and it is a useful one.

Automate the Delivery, Not the Judgment

The efficiency target one operator set for themselves is specific and aggressive: compress delivery time to 20% of what it was.

The mechanism is unglamorous. Tools like WorkBuddy handle topic selection and script drafting automatically, so the parts of a short-video commerce workflow that are mechanical stop consuming the hours that should be going into the parts that are not. A fifth of the original time is the goal, not a measured average, and I would treat it as a target someone set rather than a result you should expect in month one.

Cost is not the obstacle at this scale. Pairing WorkBuddy with BrowserAct gets a competitor pricing table in roughly 5 minutes and a product-opportunity report in about 7; a free WorkBuddy account carries 100 credits a day against roughly 10 credits per complex task, which leaves room for around 10 research runs daily before money enters the conversation at all. BrowserAct drives your real local Chrome rather than a headless browser. That matters more than it sounds, because the failure it prevents is a scrape that silently returns wrong data, and I have no idea how you would notice that before a client does.

Miora’s brand-visual template does the same job for presentation: feed it a product description and a logo, get back a consistent set of visual assets. For a small studio serving small clients, that is enough polish, and it costs minutes instead of a design retainer.

None of this touches the judgment layer. What to sell, to whom, and at what price all stay yours, and automating around them is the point.

Validate Before You Build the Product

The trap in productizing is spending two months on a course nobody wanted.

One rule I keep coming back to: spending more than two weeks choosing a direction is research used as procrastination. Give yourself three days to define a minimum sellable thing, which is probably not a product but a service, a template or a process, and then try to sell 3 of them in the first week. If they do not sell, change direction. The market answers in half a month what you would spend half a year failing to research, and honestly, the discomfort of asking three people for money is exactly what makes the test informative.

Three sales. One week. Then decide.

That rule has an unpleasant corollary for anyone sitting on a half-built toolkit, which is that the toolkit is not evidence of anything yet. Compliments are not evidence either. A payment is.

What This Realistically Looks Like

Put the pieces in order and the shape is clear enough to plan against. You start by selling units at commodity prices, because that is what is available and it teaches you the process. Somewhere in the first few hundred orders the process stabilizes, and that is the moment the SOP becomes writable. You automate the mechanical half of delivery so the same hours cover more volume. Then you sell the written process to people who are where you were six months ago, and that revenue does not scale with your calendar.

One thing the playbook posts do better than most paid courses, and I did not expect this, is the support attached to them. When someone gets stuck they post the problem and get a written answer back, and new tools get pushed out to everyone as they appear rather than being sold as a version 2. Whatever you think of the genre, that arrangement keeps people from stalling out in week two on a configuration issue nobody warned them about, which is where I would guess most beginners actually quit.

Aqiang’s 5,000 to 6,000 yuan a month is the ceiling of the first stage, not the beginning of the second. I suspect most people who quit do so while still in stage one, having concluded from a commodity price that the whole category is a commodity.

The figures in this piece all come from operators reporting on themselves, and none of them are audited. What I would take from them anyway is the sequence rather than the amounts: sell units until you know the process, write the process down, automate what is mechanical, then sell the writing. The first stage is worth 9.9 yuan a unit. The fourth one is not priced by the unit at all.

Also readable on Telegraph.

Read next

All 49 write-ups

The 2 figures in this piece — each with the sentence it came from — are in the figures table, alongside 406 more, as JSON and CSV.

Topics: Indie Development · Niche Market · Automation

Part of llm-api-pricing — field notes on AI coding agents.

Did this save you an afternoon? A star on the repository is the whole ask — it is what puts these in front of the next person looking; the data is CC BY and does not require starring.

One thing this piece could not settle: Can you reply with a yes or no: Have you ever considered selling plans instead of your time? The reply box is on the thread copy of this piece.

Want a figure that is not in here yet? Say which metric, which provider, which unit — in one line. One required field, and the page you came from is already filled in. Got a better number? Open an issue — that form knows which write-up you came from too; corrections and counter-data are the point.