AI automation agencies

The research was done by AI agents that open web pages. Some sites refuse them; where that happened we say so. The small numbers point to the source list at the end. Where a figure is our own sum or guess, it is marked “our estimate” and the basis is given.

The claim

These are examples of the claim, copied from our collection of 55 tweets about this way of making money¹. View counts are as collected on 6 October 2026. Dates are worked out from each tweet’s ID number. A ” / ” marks a line break in the tweet. We read only the text of each tweet; attached images, videos and threads were not viewed. The page judges the method, not the people who tweeted.

“I compiled 200+ n8n automation templates you can copy & paste into your business or sell to other companies.”

@damianplayer, 15 May 2025, 9,136,399 views¹

“You can make $10,000 per month with AI, If you’ve: / 1. A Laptop / 2. Wifi / 3. Time”

@Mike_Scully_, 12 May 2025, 1,687 views¹

“How to go from $0 to $10K/mo / Then $10K to $25K/mo / Then $25K to $50K/mo / All within 6 months with an AI automation agency. This is the opportunity right now.”

@lukepierceops, 2 March 2026, 4,868 views¹

“Get paid $5,000 to $15,000 to set up AI automation for businesses using a FREE tool.”

@JulianGoldieSEO, 2 February 2026, 778 views¹

The first tweet holds 84% of all the views in the collection (9,136,399 of 10,887,255)¹. Four tweets from three accounts use almost the same “copy & paste … or sell to other companies” wording, and together they hold about 86% of the views¹. So the version of the claim most people saw is the template one. Most longer tweets are cut off in our copy, including the third one above, so we quote only the part we have.

What the scheme is

An “AI automation agency” is a small business that sets up software for other businesses so that routine work runs by itself. Examples from the tweets: a chatbot that answers customer questions, a system that follows up sales enquiries by email, a voice program that answers a dentist’s phone at night¹.

The work is usually built in a “workflow” tool. A workflow is a chain of steps the software runs on its own: when an email arrives, read it, ask an AI model to draft a reply, save the details to a spreadsheet. n8n (said “n-eight-n”) is one such tool; Make and Zapier are others⁶⁷. An “AI agent” is a workflow in which an AI model decides some of the steps. A “template” is a ready-made workflow file that someone else can load and adapt.

The tweets describe three ways to earn:

1. Sell templates to companies as they are.
2. Sell the building work: charge a business once for a setup, and sometimes a monthly fee (a “retainer”) to keep it running.
3. Resell a tool under your own name. This is called “white label” (selling someone else’s software under your own brand) or “agency in a box”.

The second is an ordinary service business, and it exists. n8n’s own partner directory lists 48 firms that do this work, from small specialists to an IT consultancy that says it has over 1,200 people²⁹. The question for a reader is narrower. One tweet quoted above says a laptop, wifi and time are all you need to make $10,000 a month¹. Can someone starting from that get paying clients at the prices quoted?

The arithmetic

Step 1: what the income is made of

An agency’s income is the price per client times the number of clients, minus costs. To check “$10,000 a month” you need both figures.

The text of the tweets does not give them (attached images, videos and threads were not viewed). Of the 30 most-viewed tweets, six state a monthly income, from $10,000 to $100,000; two more give a one-off figure ($32,000 and $2.3 million)¹. In the text we collected, none gives a price per client together with a number of clients, how the clients were found, the costs or the hours worked¹.

One tweet in the top 30 gives a conversion figure. Its author says an automation “followed up with 86 leads and closed 4 deals. / $32,000 recovered”¹. That is 4.7% of leads (4 ÷ 86) and $8,000 a deal (32,000 ÷ 4), our sums. A “lead” is a person who has already shown interest. The 86 were the author’s own old enquiries, which a newcomer does not have, and the figures are unchecked.

So we take the prices from elsewhere, one route at a time.

Step 2: the price on each route

Route 1: templates. n8n’s own library holds about 13,000 workflow templates²⁵⁷. We read the library’s public data feed and counted: about nine in ten are free. 1,321 of 13,149 (about 10%) carry a price or a link to buy, set by the maker. Of the 1,311 with a price, the middle price is $20, half fall between $10 and $39, and 72 cost $100 or more⁵⁷. So a small template market does exist. But n8n publishes how often a template is viewed, not how often it is bought, so we cannot say what any maker earns⁵⁷. The most-viewed paid template has about 116,000 views at $5⁵⁷.

The tweets offer something different: 50 to 1,000 templates made by other people, to copy and sell on¹. In the text we collected, no tweet gives a price for a template or a count sold; attached images, videos and threads were not viewed¹. We found no evidence that a copied bundle sells.

Route 2a: building work found on open job boards. These are sites where a business posts a job and anyone may bid.

• On Freelancer.com on 7 October 2026 there were 11 open n8n jobs. The nine fixed-price jobs had average bids of $26 to $208; the average of those nine was $102 and the middle one $95. The two hourly jobs had average bids of $13 and $17 an hour. Each job had 12 to 379 bids, with a middle value of 68⁹. Bids are offers, not payments. This is one day on one board, where anyone may bid.
• On n8n’s own forum, a buyer asking for estimates on what he called a “small” prototype received 368 posts from 317 different people in four weeks⁴⁸. We picked out the dollar, pound and euro amounts by program. About 160 replies named one. The middle quote was somewhere between about $150 and $400, depending on whether we take the first or the largest amount in each post (the buyer also asked for running costs, so many posts hold several). Treat it as rough; it is our estimate⁴⁸.
• Other buyers on the same forum state budgets. The lowest: “a working solution for under $20”, which drew replies from 83 different people⁴⁹. Higher ones: “Budget: $1,000-2,000 for a paid first phase” for a home set-up (23 September 2026)⁵⁸; “Fixed fee per install: $300” plus $20 to $100 a month per live client, in a post that says $300 “is below the $35 to $50/hr market for this work” (16 September 2026)⁵⁹; “$1,500 USD per month” for a developer (7 June 2026)⁶⁰; and “Starting compensation: $6,000 per month” for a full-time role that is mainly coaching agency owners and asks for two years’ experience (12 August 2026)⁶¹.
• One builder on the forum advertises a fixed job at $249⁶³.

Route 2b: building work sold to your own clients. Here the highest prices come from people who sell training.

• One tweet says $5,000 to $15,000 per setup¹.
• A course seller’s 2024 article prices five example systems at $1,500 to $10,000 each⁵⁶.
• A training firm’s totals (Step 5) divide out to about $18,000 per tracked client¹². That is our division of the firm’s own unaudited numbers.
• Reviewers and testimonials for the same firm mention a first client at $3,000, a $10,000 deal and “$20k worth of AI consultancy work” (see “What people who tried it report”)¹⁴¹⁷.
• Two sets of prices come from people who do not sell training. Rezora, a firm that set up AI phone calling for real-estate brokers, says it charged “an initial thousand to $2,500 setup fee”, “plus an ongoing $500 month monthly fee, plus 20 cents per conversational minute”⁷⁴. That is the founders’ own statement.
• n8n’s partner directory lets each listed firm say which project budgets it accepts, in six bands from “$250 - $1,000” to ”+$100,000”²⁹. We opened 45 partner pages, and 43 show bands. The lowest band accepted is $250 to $1,000 for 1 firm, $1,000 to $5,000 for 17, $5,000 to $10,000 for 16, and $10,000 to $80,000 for 9 (our count)⁷⁰. So 42 of the 43 say they take nothing under $1,000. These are bands the firms say they accept, not payments. The firms are established: n8n lists an agency only if it has “at least three customers actively using n8n”²⁹.

We found no independent survey of what a small business typically pays for an automation. What we have is a spread: tens to a few hundred dollars for small jobs on open boards, $1,000 to $2,000 for a scoped project posted by a buyer, $1,000 or more as the stated floor at 42 of 43 established n8n partners, and $1,500 to $15,000 from sellers of training and their members. The gap is the heart of the matter. It is the difference between bidding against hundreds of strangers and having a business owner who already trusts you.

Route 3: reselling a tool. This has a cost before any income (Step 6), and it is where a regulator has acted (see “What the rules allow now”).

Step 3: the volume needed

Jobs needed each month = income wanted ÷ what you keep per job. These are our sums. Freelancer.com takes “10% or $5.00 USD, whichever is greater” of each project¹⁰, so a $102 job leaves about $92. Fees on the other rows are ignored. Where one sale is worth more than $1,000, the table says how many months it covers.

To earn $1,000 a month from Freelancer.com at those bids you would need to win about 11 jobs, each against a middle value of 68 other bidders⁹. We do not know how many n8n jobs are posted there in a month; 11 were open on the day we looked. At the hourly bids of $13 to $17, $10,000 a month needs roughly 650 to 850 paid hours after the fee; a month has about 730 hours in total (our sums from⁹¹⁰). At the $35 to $50 an hour that one forum buyer calls the market rate, it needs 200 to 286 paid hours a month (our sum)⁵⁹. A full-time month is about 170 hours (our estimate: 40 hours times 4.3 weeks).

At the sellers’ prices the volume looks easy: one or two clients a month. So the claim does not rest on volume. It rests on whether a newcomer can get even one client at that price.

Step 4: finding one client

Cold email means writing to businesses that have not asked to hear from you. Several tweets promote it (“Auto-Send Cold Emails While You Sleep”)¹. Instantly, a company that sells cold-email software, published a benchmark for 2025: “The overall average reply rate is 3.43% with top-performers of users exceeding 10% reply rates”⁸. The report counts replies of any kind, including “no thanks”, and gives no figure for meetings or sales⁸. The source sells the method, so it has a reason to show it working. Its software costs $47 a month for 5,000 emails⁸.

No source we found measures how many replies become paying clients, so we do not put a number on it. One account is in the reviews below: a software engineer who says he sent about 240 emails a day and earned nothing; the company disputes his account¹⁶.

The competition for each job is measurable. A middle value of 68 bids per job on Freelancer.com⁹. 317 people answering one forum request, where the buyer asked for “One relevant project you personally delivered, with a demo if available”⁴⁸. A newcomer has no such project. That thread was the busiest. Across the 18 hiring threads opened on the forum’s first page since 7 September 2026, the middle one had about 48 replies, and the range was 1 to 368 (our count; replies are not the same as separate applicants)⁸⁶. And the free community attached to one training programme has about 335,600 members, all being shown the same business¹².

One tweet says “learn workflow automation now and customers will find you”¹. Nothing we read supports that. Other tweets in our collection say the opposite. One says of the label “AI automation agency”: “the market is flooded with these non specialized agencies which is why you see a race to the bottom on pricing”¹. Another passes on a speaker’s point from a video clip, which we did not watch: an offer to automate a whole business is too vague for a buyer to act on¹.

Step 5: how many get there

No study, platform or regulator publishes what share of new AI automation agencies get a paying client. The only group figures come from a seller of training.

Morningside Education runs a free community and a paid programme, the AAA Accelerator¹²¹³. A 2023 tweet by @liamottley_ in our collection advertises one-to-one help with starting an “AAA”; we did not open its link, so the match to this programme is our inference¹. The community’s page says: “the 7,250+ agencies we’ve trained through the AAA Accelerator over 4 years: 1,000+ clients landed, $18M+ in client revenue generated for our members.”¹²

Our sums from those figures:

• 1,000 clients ÷ 7,250 agencies = about 0.14 tracked clients per agency trained.
• $18 million ÷ 7,250 = about $2,500 of client revenue per agency over four years.
• $18 million ÷ 1,000 = about $18,000 per client.

This ratio is not a success rate, and it can be wrong in either direction. The numbers are the firm’s own and unaudited. The ”+” means they are minimums. The firm’s sales page calls the $18 million “tracked deals signed”, so deals that members did not report are missing, and the true totals could be higher¹³. The 7,250 counts everyone trained over four years, including people who joined last month; the sales page puts the current community at “4,000+ agency owners”¹³. And clients are not spread evenly. The firm’s case studies show single members with several clients each, such as “3 full AI system builds signed”¹⁴. So the share of members with at least one tracked client is lower than the ratio suggests. What the figures do show: the firm tracks far fewer clients than agencies trained, and the clients it tracks paid well.

The sales page gives slightly different figures, which may simply be older: 7,150+ members, 800+ clients, $18M+. It also shows “$32K AVG MRR” and “On average, our members land their first client in 41 days.”¹³ MRR means monthly recurring revenue, the income that repeats each month. If all 7,150 members averaged $32,000 a month, that would be about $229 million a month (our multiplication), which cannot fit an $18 million total. So the average must describe some smaller group, and the page does not say which. The 41 days, likewise, can only cover members who landed a client. In a public reply to a reviewer in July 2026, the firm gave that average as 61 days¹⁶.

As a general yardstick, US government data show 77.9% of business establishments opened in the year to March 2024 were still open a year later²². That covers all industries and only businesses with employees. There is no such figure for this scheme.

Step 6: the costs

Costs are low. That is also why so many people enter.

• The workflow tool. n8n’s hosted version is 20 euros a month (2,500 runs) or 50 euros (10,000 runs), billed yearly. A version you run on your own server (“self-hosted”) is free to use⁶. Zapier starts free, then from $19.99 a month billed yearly; paying month by month costs more (at its 2,000-task level, $73.50 against $49)⁷. Make starts free, then $9 a month; its page offers a discount for paying yearly and we could not tell which billing the $9 belongs to⁷. A basic rented server at DigitalOcean, one hosting company, is listed at $4, $6 or $12 a month, depending on memory⁸⁹. So running n8n yourself costs less money than the hosted plan and more upkeep. We did not find the smallest server n8n needs, and did not test one.
• The AI model. OpenAI’s newest family of models costs $0.10 to $10 per million “tokens” (pieces of words) read, and $0.50 to $50 per million written; older and specialist models run from $0.05 to $150 read²¹. For a small business’s volume this is usually minor.
• Voice agents. One tweet recommends selling AI voice agents instead¹. These cost money per minute of call: about $0.08 to $0.13 at Vapi before phone charges (our sum of its listed parts), and “$0.07-$0.31 / min” at Retell¹⁹. A client with 1,000 minutes of calls a month costs the agency roughly $80 to $310 (our multiplication).
• Agency-in-a-box software. HighLevel costs $97 a month for 3 client accounts, $297 for unlimited accounts and $497 for the plan that lets you resell the software at your own prices²⁰. That is owed every month before the first client.
• A typical starting bill (our estimate). n8n’s hosted Starter plan (20 euros), a cold-email plan ($47⁸) and one paid community ($69¹⁸) come to about $140 a month, taking the euro at $1.13, the European Central Bank’s rate for 6 October 2026⁶².
• Tax. In the United States, “The self-employment tax rate is 15.3%”, on net earnings from $400, on top of income tax⁴⁰. The incomes in tweets are before this.
• Training. This is the large cost; see “Who makes money from it”.

What people who tried it report

A limit first. Web search was not available to us when this page was first researched and reviewed, so every page was opened by its address. A later pass on 7 October 2026 had 27 web searches. That is a small search. It found the success accounts now set out under “The upside”, all of them self-reported. Reddit, Upwork and Fiverr still refused us. So the lack of independently confirmed incomes here covers only what we could search and open.

Income figures in the tweets. We read only the text of each tweet. Attached images, videos and threads were not viewed, so we cannot say what proof any tweet carries. Three tweets point to interviews by a third person. One interviewer writes of a “$1+ million AI automation agency” built largely on Zapier: “I was skeptical. / So he screenshared and showed me what his clients pay him for.”¹ The others cover a “$20,000 per month” agency and “Zak went from 0 to $12k/mo”¹. We did not watch these videos, so we can neither confirm them nor count them as unproven. They are the likeliest place to find documented cases.

The “$1+ million” story appears to match a man interviewed by Zapier’s chief executive on Zapier’s own podcast in September 2025. The match is our inference from the details of the story, and we did not find the interviewer’s own video. The transcript gives no income figure. It does say what the path was: he sold roofing door to door, built automations inside a roofing company where he worked, did free work until people offered to pay (“please let me do free work. You don’t have to pay me. And at the end of the day, these guys were like, please let me pay you.”), took a job at a software company, and left when his own business outgrew the job. That took about four years⁸⁰.

We found no public page where a payment company confirms an AI automation agency’s income. The later pass looked again. We scanned the names of the 11,378 listings on TrustMRR, a site that shows revenue confirmed through payment companies such as Stripe, for words such as agency, automation, n8n, workflow and consulting. The automation-service listings we found showed no confirmed revenue or a few hundred dollars a month; the highest was $674 of monthly recurring revenue⁷⁶. Agencies in other fields show far more there, such as a Reddit marketing agency at $86,000 a month⁷⁶. This is weak evidence. It is a match on names only, the site lists mostly software, and agencies often bill by bank transfer, which such a site does not see.

Income surveys that should not be relied on. Search results on this subject are led by articles from one site that claim a survey of 81 freelancers with a middle income of $94,000 a year. The article publishes no method and no data, and says all 81 “run at least $3K/month in automation revenue”. A sample that leaves out everyone below $3,000 a month cannot show a typical result, so we do not use it⁷⁸.

Reviews from people who paid for training. These are from the Trustpilot page for the AAA Accelerator, read in full through a reader tool¹⁶¹⁷. Reviews are self-reported. The company answers most low-star reviews in public, and we give its reply where it disputes the account.

• A two-star review titled “Good coaching, but I made $0” comes from a writer who says he is an experienced software engineer and was a member for about five months. He writes: “I made $0. Not because I didn’t know how to code or use AI. Not because I didn’t work hard.” and “the business model is not easy money”. He adds that “the accelerator does offer structure and good coaching” (shown with the date 21 January 2025)¹⁶. The company replied that “call transcripts show a gap between the strategies we recommended and what was actually implemented” and that he had “roughly three active months”. He disputes that reply in an update¹⁶.
• One reviewer reports paying £5,000. Another reports 6,300 euros in total, then 250 euros a month to continue¹⁶.
• A three-star reviewer got his first paying clients during his six months and calls the programme “worth it”. He warns: “Just don’t believe sales people that say you at least hit 5k-10k dollar assignments within the first 6 months without too much prior experience.” (25 February 2025)¹⁶
• A member of more than a year (14 months, according to the company’s reply) writes: “This is not for ppl trying to start a company. This is for ppl who already have a business and expertise in an industry” (13 July 2026). The company replied that he missed a third of the coaching sessions he booked and left with “a running outreach business”¹⁶.

Positive reviews, and what they measure. The programme scores 4.7 out of 5 from 548 reviews, 92% of them five-star¹⁷. Some report income. One writes: “I was able to get $20k worth of AI consultancy work” (6 May 2026). Another: “First client, first build and really great start for my business (with a great onboarding fee and retainer in place)” (22 May 2026)¹⁷. Across the roughly 40 most recent five-star reviews, by our count four mention a client, a build or a deal and one gives an amount; most describe coaching calls or the first weeks of membership¹⁷. So the rating mostly measures satisfaction with the coaching, not income.

The seller’s own case studies. The programme’s case-study page shows eight named people: one member story and seven community posts¹⁴. Two give an agency revenue figure: “$255K+ in 65 days” and “2K to 25K/mo”. One gives a client’s result (a law firm helped to “recover $1.6M in lost revenue”), one a “$50K first VC investment”, and four give no money figure¹⁴. The same page displays reviews with smaller figures, “First $3k client within 3 weeks” and “Closed a $10,000 AI consulting deal”, which are closer to what a newcomer might see¹⁴. None shows payment proof such as an invoice. The programme’s terms say: “Individual outcomes depend on a wide range of personal factors, and no specific results are guaranteed.”¹⁵

One well-known operator publishes a figure for his teaching community and a figure for his earlier service businesses. Nick Saraev, who is not in our tweet collection, runs a paid community on Skool, a site that hosts such communities. His own site says he co-founded “two service businesses” that “scaled to a combined $160,000/mo in revenue”, and that his community is “the largest AI community on Skool by revenue ($330,000/mo)”⁵². The $160,000 is an undated past figure for businesses he co-founded; his site names them as LeftClick and 1SecondCopy, and we did not check what each sold. The two figures may not be from the same period. All of it is self-reported. He did not start with automation. A profile by Apify, a tool company that pays him for referrals, says that in 2019 he “launched a door-to-door digital marketing company, connecting with 50 to 100 local businesses each day”, and later built a content business to $90,000 a month before the automation agency and the community⁹¹. Those figures are his own too.

Demand does exist.

• The first page of the Jobs section of n8n’s forum showed 19 threads opened in the three weeks to 6 October 2026. About 15 were from someone looking to hire for n8n work; the rest were builders offering services or seeking a sales partner, or not about n8n (our count of titles)⁴⁷.
• Upwork, the largest freelance marketplace, reports client spending in its results filed with the US securities regulator. It says spending on AI-related work “surpassed $300 million on an annualized basis in Q4 2025, up more than 50% from the prior year”, and that spending on “AI Integration & Automation”, its largest AI category, “grew more than 90% year over year in Q4 2025”⁶⁴. This is spending across the whole site, not anyone’s income.
• That growth has slowed. AI Integration & Automation grew “more than 50%” in the first quarter of 2026. For the second quarter Upwork gave no figure for that category, said all AI-related work grew “more than 22%”, and reported total spending on the site of $966.4 million, down 4% on a year before⁶⁵.
• On Fiverr, another marketplace, larger jobs are growing while the whole is shrinking. Projects of $1,000 or more were 15% of the value of completed projects and grew 13% in a year, and programming and technology projects above $1,000 grew 34%, while Fiverr’s total revenue fell 10%⁷⁷. We read this in a summary of Fiverr’s July 2026 results, not Fiverr’s own release.
• A study from MIT’s NANDA project, dated July 2025, found most organisations saw no measurable return from AI projects. It also found “external partnerships with learning-capable, customized tools reached deployment ~67% of the time, compared to ~33% for internally built tools”²⁸. The sample leans to large companies. It cuts both ways: outside help does better, and most projects return nothing, which makes clients hard to keep.
• The US Census Bureau reported in May 2026 that 17% to 20% of businesses use AI, and “Less than 20% of firms with four or fewer employees reported using AI.”⁵⁵ That leaves many possible clients, or many firms that do not want it; the figure cannot tell these apart.

What the rules allow now

Building for clients is allowed. n8n’s licence page says you may “Charge your customers for workflow creation, setup, and maintenance services”, as long as “clients receive outputs, not access”⁴. It sets no limit on the number of clients⁴.

Renting out the tool is not. The same page lists what the free licence does not allow, including “Host n8n as a service and allow your clients to build workflows” and “White-label n8n by removing our branding and offering n8n as is.”⁴ The licence text limits the free version to “your own internal business purposes”⁵. Other uses need a paid agreement; we found no price for it.

Selling templates. Makers sell their own templates through n8n’s library, so that is plainly accepted⁵⁷. The licence pages cover the software and say nothing about workflow files⁴⁵. We did not check what rights you have to resell a template someone else made.

Running your own server means looking after it. n8n published ten high-severity security notices, and four moderate ones, on one day, 30 September 2026³¹. An agency that keeps clients’ passwords on its own server has to apply such fixes. That is more than the laptop, wifi and time of the tweet quoted above¹.

The client-finding methods in the tweets run into rules.

• Collecting contacts by scraping. A tweet promises to “Scrape 1,000s leads in 1 click” and, in the text we collected, does not say from where¹. If the source were LinkedIn, its user agreement forbids using “bots or other unauthorized automated methods to access the Services, add or download contacts, send or redirect messages”³².
• Cold email in the United States. It is legal if done by the rules. The Federal Trade Commission (FTC), the US consumer protection regulator, says of the CAN-SPAM Act: “The law makes no exception for business-to-business email.” Each email that breaks it can cost up to $53,088, and both the business and the firm sending for it can be liable³³.
• Gmail’s limits. Google requires senders to keep spam complaints under 0.3%³⁴.
• n8n’s hosted service. Its acceptable use policy bans spam and allows n8n to “temporarily suspend access to the Service, with or without prior notice”³⁰. A suspended account stops every client workflow on it.

AI voice calls need consent in the US. The Federal Communications Commission (FCC) ruled in February 2024 that “callers must obtain prior express consent from the called party before making a call that utilizes artificial or prerecorded voice simulated or generated through AI technology”³⁵. This covers calls the AI places, not calls a customer makes to the business.

Chatbots must say they are AI, and some subjects need a professional. Anthropic’s policy: “All consumer-facing chatbots, including any external-facing or interactive AI agent, must disclose to users that they are interacting with AI rather than a human.” It also requires a qualified professional to review output in health, legal and financial uses³⁶. OpenAI’s policy bars “provision of tailored advice that requires a license, such as legal or medical advice, without appropriate involvement by a licensed professional”³⁷. Article 50 of the EU AI Act sets a similar duty to tell people; we read it on an unofficial copy³⁹. A tweet suggests selling AI agents to dentists¹; rules on patient data would apply, and we could not open the official page.

Platforms can close a product. Meta’s terms for the WhatsApp Business Platform, dated 23 September 2026, say AI providers whose AI is the main thing offered “are strictly prohibited from accessing or using the WhatsApp Business Platform”³⁸. We read the scope only in summary.

The closest regulator case is about selling the opportunity. In August 2025 the FTC sued Air AI. The complaint says the company sold coaching and software “for prices ranging from $15,000 to $30,000”, and licences to resell its AI to other businesses “with prices ranging from $25,000 to $100,000”, and took roughly $19 million from buyers²³. It says: “many consumers who purchase the Licensing Business Opportunity are not able to resell it to anyone”²³. The FTC’s announcement says buyers “lose as much as $250,000 and are often left in debt after relying on Air AI’s false promises”²⁴. Paragraph 10 of the complaint we read says “as much as $100,000”²³. The complaint’s full text says $100,000 several times and never $250,000, so the larger figure is the announcement’s alone²³²⁴.

These were allegations, and the case was settled without a trial. The settlement announced on 24 March 2026 carries an $18 million judgment, mostly suspended because the defendants could not pay, with $50,000 to be paid, and bans the company and its owners from marketing business opportunities²⁵. The court has approved it. The judge signed the order on 30 March 2026 and it was filed the next day. It enters “Judgment in the amount of Eighteen Million Dollars ($18,000,000)”, orders the defendants “to pay to the Commission $50,000” and suspends the rest, and bans them for good from selling any business opportunity. The defendants neither admit nor deny the allegations⁹⁰. The case concerns one company’s licences. It is not a finding about building automations for clients.

These cases concern the companies named in them only. We found no regulator action against any account quoted or named on this page, and we do not suggest any connection.

A check a reader can use. The FTC’s Business Opportunity Rule requires sellers it covers to give a disclosure document seven days before taking money. If they make an income claim, they must state in writing the “number and percentage of your buyers who got at least that result”²⁷. Whether the rule covers a particular course is a legal question we did not settle, and we make no finding about any named seller. But anyone can ask a seller for that number. The FTC said in 2024, announcing unrelated cases about AI-related marketing claims, that “there is no AI exemption from the laws on the books”²⁶.

Who makes money from it

In our 55 tweets. By our count from the collection’s labels, about 45 of the 55 point the reader to something of the author’s: a course, a paid community, a “reply and I’ll send it” giveaway, a tool, a video channel or the author’s own services¹. About 19 lead to a course, community or coaching, against 8 that lead to the author’s own agency work¹. The labels are our one-line reading of each tweet’s text and profile; links were not opened, so the counts are approximate.

Community owners. One account, @JulianGoldieSEO, wrote 11 of the 55 tweets¹. One of them names a paid community, AI Profit Boardroom, and three more carry the same short link, which we did not open. Its page shows 3,406 members and “JOIN $69/month”; a tweet in our collection gave $59¹¹⁸. Our estimate: 3,406 x $69 = at most about $235,000 a month. That is a ceiling, not his income: it assumes every listed member pays the current price, and the page does not say how many do. The page says “Boardroom members are PROFITING from my personal $300K/month AI tech stack”; the page as we read it on 7 October 2026 shows no figures for members’ earnings¹⁸. It offers a “7-Day No-Questions Refund” and a “30-Day ROI Guarantee” (a refund if one workflow does not save time or bring leads or revenue); we tested neither¹⁸.

A second community, Maker School, charges $184 a month and shows about 2,500 members. It promises: “Get your first paying AI client in 90d or your money back.”⁵⁴ The page as we read it on 7 October 2026 gives no count of members who landed a client and no conditions for the guarantee⁵⁴. Its owner’s site says it brings in $330,000 a month and has “10,000+ graduates”⁵². If “graduates” means past members of this community, about four times as many people have been through it as are members now; the site does not define the word. Neither page publishes a count of first clients or of refunds.

Training programmes. The AAA Accelerator shows no price and asks visitors to book a sales call¹³. The only prices we have are the reviewers’: £5,000, about $5,000, and 6,300 euros plus 250 euros a month¹⁶. The firm behind it also runs a client business, Morningside AI, whose site lists named corporate clients¹²⁵⁰.

Audience builders. The most-viewed tweet asks readers to “Follow + Reply” to receive the templates¹. A reply-to-receive post can grow its author’s following whether or not any reader sells an automation.

Tool companies pay for referrals. A referral or “affiliate” payment goes to whoever sent a new customer.

n8n also tells template makers: “Submit your template to the n8n template library, get featured, and earn via our affiliates program.”² On n8n’s 20-euro plan, 30% is about 6 euros a month per referred customer for a year (our sum)³⁶. So a free-template post can pay its author without anyone selling an automation. We do not know whether any author in our collection is an affiliate; the links were not opened.

Platforms. Skool charges community owners $9 a month plus 10% of member payments, or $99 a month plus 2.9%⁴². Whop, named in one tweet, takes “2.7% + $0.30 per successful transaction for domestic cards”⁴⁶. Both are paid whether or not members earn anything.

People who do the work. Some do. One tweet says, in capitals, that a free tool called n8n-MCP is “WHAT EVERY $10K/MO AI AUTOMATION AGENCY IS RUNNING BEHIND THE SCENES”¹. Its developer writes that maintaining it “competes with my paid work”⁵¹. He is a software engineer with public work that anyone can inspect. That is a different starting point from a folder of copied templates. It has a visible result: he is listed in n8n’s partner directory, where his page says he accepts projects from $1,000 to $80,000 and shows 3 client reviews⁷¹. The page gives no income figure.

The upside

Some people do earn from this work. This section sets out the best outcomes we could document and how each one is known.

One limit applies to all of it. We found no income figure for an AI automation agency that a payment company, an auditor or a journalist who checked records has confirmed⁷⁶. Every income figure below is the person’s own statement. What third parties do record is counts of client reviews, company filings and total spending on marketplaces.

The best documented outcomes.

The interviews with Rezora, PressW and the consulting firm were sponsored by Zapier, which sells automation software⁷²⁷⁴⁷⁹.

The best-documented path is a modest one. Andrew Davison says he learned Zapier for his own small business, joined Upwork, was freelancing full-time “by 2018” and formed a company in 2019. His background was sales, not software. He says: “In the early days, new work all came from Upwork or Fiverr, but we now get most of our clients from the Zapier Experts program”⁶⁸. Zapier’s directory lists him as a Gold partner with a company size of 1, a rating of 5.0 from 239 reviews, and a minimum spend of $100⁶⁷. The reviews are a platform’s record of client work, not an income figure.

UK company filings show how long it has lasted. Luhhu Ltd, formed on 11 December 2018 with him as director, is still active. Its accounts, filed every year, report 0 employees⁶⁹. Accounts for a company this small do not show sales, profit or the owner’s pay, so they neither confirm nor contradict his monthly figure. They show a small one-person business that has lasted more than seven years.

The larger figures are a different kind of business. PressW calls itself an AI engineering firm and builds custom software. Its founder had been chief technology officer of an AI product company that was sold, and he started with his two founding engineers⁷². The consulting firm sells training days at $15,000 and contracts at “15 grand a month” to established companies. Its founder had built and sold a startup backed by investors⁷⁹. Neither sells templates or simple workflows.

One narrow problem in a known industry. Rezora’s founder was a real-estate broker with a computer science degree. He built an AI calling system for himself and then sold it to other brokers. The founders say the first customer came less than four days after they started Facebook adverts⁷⁴. He then stopped the agency, because more than 40 clients each wanted changes, and turned the service into software⁷⁴. So in this case the agency was a stepping stone.

What the top tenth looks like. No source gives incomes across this field, so we cannot say what the top tenth or top hundredth earn in dollars. One stand-in exists. Zapier’s partner directory lists 556 firms and freelancers and shows a count of client reviews for each⁶⁶. We captured 402 of the 556; the page changed the order of entries between loads, so we missed about 150. Of the 402:

• 272 show no reviews.
• 130 have at least one, 59 have 10 or more, and 16 have 50 or more.
• 9 have 100 or more. The highest is 473.

So the middle partner in our capture shows no reviews, and a partner needs about 17 to be in the top tenth. The top tenth holds about 90% of all the reviews. These are our counts, and they are partial⁶⁶. Even the partner with the most reviews accepts small jobs: Flow Digital, with 473 reviews and 16 to 50 employees, lists a minimum of $100 and sells a 75-minute session for $297⁷⁵.

A realistic good result in the first year or two. We cannot give a typical income, because no source measures one. What was observed:

• First jobs on open boards at about $100 to $400, and up to $1,000 to $2,000 for scoped projects posted by buyers⁹⁴⁸⁵⁸.
• One reviewer got first paying clients within a six-month programme. One skilled reviewer made nothing. Some five-star reviewers report a first client at $3,000 or deals of $10,000 to $20,000¹⁴¹⁶¹⁷.
• One anonymous agency owner’s account of about a year: roughly 20 workflows sold and one long-term client still paying. He writes: “I made money, but the more clients I had, the more software I had to maintain.” We read this on a copy of a Reddit post, not on Reddit⁸².
• In an undated interview given between 2019 and 2022, the Zapier practice above said the business was making $5,000 to $8,000 a month⁶⁸. He had gone full-time in 2018. That is what the business took, not his pay, and it dates from before the current interest in AI.

$10,000 a month or more is claimed by some; we could not open any proof of it. Our judgement on this evidence: a good first year is a handful of paid jobs and perhaps one client who keeps paying. After several years the best-documented result is a small one-person business that has lasted: 239 client reviews and seven years of company filings; its income is the owner’s own figure, which filings for a company that small do not show⁶⁷⁶⁹.

Too early to tell, or nobody wants it? Both readings fit the low share of small firms using AI⁵⁵. The marketplace figures lean towards “early”, on our reading: spending on this work at Upwork grew fast in 2025 and more slowly in 2026⁶⁴⁶⁵. What tells the two apart for you is the price business owners near you will pay for one fixed problem. A few conversations will show it, at no cost, before you buy anything.

What it takes to compete

These points are our reading of the sources above, not a measured result. No source measures which of them matters most.

What the people who got there had.

1. They could build, or spent years learning. Real client work means connecting business systems, handling errors and keeping a server patched³¹. PressW began with a former chief technology officer and two engineers, and its founder says almost 80% of the team are engineers⁷². Rezora’s founder has a computer science degree⁷⁴. The man on Zapier’s podcast took about four years to go from his first automation to his own business⁸⁰. Skill alone is not enough: the engineer who reported earning $0 had it¹⁶.
2. They knew an industry before they knew the tool. A roofing salesman automated the roofing firm he worked for⁸⁰. Zapier’s chief executive gives a roofer who automated his own office and then “set up a, uh, a roofing automation agency where he works with roofers” as his example of a business built on Zapier⁸¹. A broker sold to brokers⁷⁴. The long-term member’s review says the training suits “ppl who already have a business and expertise in an industry”¹⁶. Tweets in our own collection advise starting with local business owners¹. The one-person Zapier practice is an exception: it found clients on marketplaces and then through Zapier’s directory⁶⁸.
3. The first client came from someone who knew their work. The rest came from contacts they built. PressW’s first customer was a partner’s former employer. Its founder says: “It’s relationships and networks.” But he also says: “Less than 10% came from like, old relationships. I’d say 90% were people that I’ve met and interacted with since starting the business.” He meets them at business events, including ones he hosts; one cost him about $15,000, and he says one deal from it paid that back⁷². The consulting founder’s first customer was a local chief executive he half-knew, and he had posted on LinkedIn daily for years⁷⁹. The accounts say contacts can be built while doing the work. What nobody describes is succeeding without them.
4. They sold one specific result. One tweet puts it as “one buyer / one painful workflow”¹. The anonymous one-year account lists what he learned: “No niche = chaos”, “One-off automations = disappointment” and “Long-term engagement works better”⁸².
5. Their clients came from outside the bidding boards, or they used the boards only to start. On the boards we saw bids and quotes of roughly $100 to $400 a job against dozens or hundreds of bidders, and buyer budgets up to $1,000 to $2,000 for a scoped project⁹⁴⁸⁵⁸. The higher prices all assume a direct relationship¹⁵⁶. Andrew Davison began on Upwork and Fiverr, collected reviews, and then moved to Zapier’s partner directory for most of his clients⁶⁸.
6. They could cover months with little income. One skilled reviewer reports no income after five months, three of them active by the company’s count¹⁶. The one-year account says a real automation takes three to six months inside a client’s company⁸².
7. They charged for upkeep. Two accounts say looking after delivered work is what stopped them growing: the one-year account calls it a “bottleneck”, and Rezora’s founder stopped taking agency clients over it⁷⁴⁸².
8. They sold their work, not the tool or other people’s templates. The licence forbids renting out n8n⁴. Nine in ten templates are free at source and the rest are priced near $20⁵⁷. The one “resell our AI” offer for which we found a regulator’s case ended in a settlement, with no admission, that bans the sellers from marketing business opportunities²³²⁵.

What buyers ask you to show. One buyer asked for “One relevant project you personally delivered, with a demo if available”⁴⁸. Two more hiring posts from September and October 2026 ask for “2-3 examples of workflows you have built” and “Examples of similar work”⁸⁸. One of those came from someone who had already built AI agents and wanted a second person to “test edge cases and error handling” and fix “anything that isn’t client-ready” before showing clients⁸⁸. So a demo that runs is not yet work a client can rely on.

What it costs.

• Money: little. Tools cost about $140 a month on our estimate (Step 6), and a rented server $4 to $12 a month⁸⁹.
• Training: optional and large. Reviewers of one programme report paying thousands¹⁶.
• Time: the larger cost. The man on Zapier’s podcast took about four years to reach his own business⁸⁰. The one-year account ended that year with one long-term client⁸².
• Proof. n8n will list an agency as a partner only if it has “at least three customers actively using n8n” and automation is its main business, and the programme is a closed pilot open only in some regions²⁹. That is one checkable line between starting out and established.

How crowded it is. The number of people using the tool is rising fast. n8n’s forum has 180,338 registered users and added 5,078 in the last 30 days⁸³. n8n’s code project had “more than 70,000” stars (a rough count of interested developers) in March 2025 and 206,807 on 7 October 2026⁸³⁸⁵. n8n said in October 2025 that it had seen “6x user growth, 10x revenue growth” in a year⁸⁴. These figures cannot separate rival builders from businesses that now build for themselves. Both make the work harder to sell.

Some of the demand is other newcomers. Several of the forum’s hiring posts come from new agencies looking for someone to do the building. One says: “We’re starting from zero, with no existing clients or fixed monthly salary.” It drew replies from about 48 people⁸⁷. Builders are plentiful there. Clients are the scarce thing.

How to tell early which side you are on. These checks are our judgement.

• Can you name five business owners who would take your call? If not, your route is cold email or the bidding boards, where the measured numbers are weakest⁸⁹. Or it is building contacts first, which the accounts above say takes time and sometimes money⁷².
• Has anyone paid you anything in the first three months? A first job at $200 is a better sign than a finished course. So is someone offering to pay after seeing free work⁸⁰.
• Do you have one working example in one industry that you built yourself? Buyers ask for it⁴⁸⁸⁸.
• What have you spent against what you have been paid? Tools cost about $140 a month on our estimate. Training reported by reviewers costs thousands¹⁶.
• Does the plan need you to buy something first? Ask the seller what number and share of buyers reached the income advertised²⁷. No seller page we opened gives it.
• Is the thing you plan to sell a bundle of other people’s templates, or access to the tool? Then expect about $20 a sale at best, or a licence problem⁴⁵⁷.

What we did not verify

• How many people who try this get a client. No independent figure exists that we could find. The only group numbers are one seller’s own, unaudited, and count tracked deals. Our per-agency sums are not a success rate.
• Search was thin. It was not available in the first research or its review. A later pass used 27 searches. One search for journalism that checked the records of these courses or their students found nothing relevant; that is one search, not proof that none exists. We did not search for Google Trends data or for regulator actions outside the United States.
• Every success figure. The incomes reported for Luhhu, Rezora, PressW, the consulting firm, Nick Saraev’s businesses and Morningside’s members are all their own statements. We found no filing, payment page or record-checked article for any of them. Luhhu Ltd’s accounts do not show sales, profit or the owner’s pay. The Luhhu interview is undated. Nick Saraev’s background comes from a profile written by a company that pays him for referrals.
• What the top earns. No income figures exist across this field. The Zapier review counts stand in for the amount of client work, not income. Our count covers 402 of 556 partners. We did not check how Zapier collects or checks reviews, or what each partner level requires.
• How long it takes. No source measures the time to a first client or to a living income. Only single accounts exist.
• The n8n partner budgets. Our count covers the 43 of 45 opened pages that showed a budget band; the directory lists 48 firms. An earlier count of ours found bands on only 30 pages, so it missed 13. The bands are what firms say they accept, not what clients paid.
• The income claims in the tweets. None was checked. We read only the text of each tweet; attached images, videos and threads were not viewed, and links were not opened. We did not watch the linked videos, including three third-party interviews. The match between the “$1+ million” tweet and the Zapier podcast guest is our inference, and that transcript gives no income. Most longer tweets are cut off at about 280 characters in our copy, including several quoted here. Tweet dates are worked out from ID numbers.
• Upwork, Fiverr and Reddit. All refused us, so freelancers’ profiles, rates and earnings on the largest marketplaces are missing. We read Upwork’s totals in its filings, Fiverr’s in a summary, and one Reddit account on a third-party copy whose author is anonymous. A reported Upwork release on which skills are most in demand could not be opened, so we do not use its figures. The Freelancer.com data is one day on one site with 11 jobs.
• What a small business really pays. The prices above are a tweet’s claim, a course seller’s article, our division of a seller’s totals, reviewers’ own accounts and single forum posts.
• Template sales. n8n shows prices and views, not sales. We do not know what any maker earns, or whether anyone buys copied bundles.
• How often a cold-email reply becomes a client. No source.
• The forum price figures. A rough count by program across mixed currencies. We do not know who, if anyone, was hired or paid, or whether the reply counts equal separate applicants.
• Exact quotes. Some pages were read through a tool that extracts or summarises text. Prices for n8n’s hosted plans, Make, Vapi, Retell, HighLevel and OpenAI and the BLS figures were not re-checked in the raw page. Make’s and Zapier’s month-by-month prices at the lowest level were not seen.
• Trustpilot. The site shows one date per review, which may be the date of the experience and not of posting. Review counts and star shares were read once.
• Sellers’ terms. The AAA Accelerator’s price, refund terms and contract are not published. No refund guarantee was tested. Owners’ income claims and our members-times-price estimate are unchecked.
• The Business Opportunity Rule. We read the FTC’s 2011 guide, not the current rule text, and did not settle whether it covers any seller here.
• Rules not read. Patient-data rules; cold-email law in the EU and UK; state laws on AI disclosure and telemarketing; US text-message registration; other tools’ terms on reselling; X’s rules on reply-to-receive giveaways; whether the EU AI Act date was moved.
• Other prices. n8n’s licence for embedding the tool, the voice tool named in one tweet, and the smallest server n8n needs. The server prices are list prices from one host, not a tested set-up.
• The MIT report was read on a third-party copy.
• Whether the people named have seen this page. They were not contacted before publication.

Sources

“Read at source” means we opened the page itself on 7 October 2026, in many cases through a tool that extracts the text. “Secondary” means we read a copy or summary, not the original. “Estimate” means our own sum.

1. Does It Pay collection of 55 tweets on this scheme, collected 2026-10-06; individual tweets linked above. Read from our collected copy on 2026-10-07, not from the live tweets. Counts by label are estimates.
2. n8n, workflow template library. https://n8n.io/workflows/ . Undated live page showing 13,054 templates. Read at source on 2026-10-07.
3. n8n, affiliate programme. https://n8n.io/affiliates/ . Undated page. Read at source on 2026-10-07. The 6 euros figure is our estimate.
4. n8n Docs, community licence FAQ. https://docs.n8n.io/n8n-community-license/community-license/license-faq . Undated page. Read at source on 2026-10-07.
5. n8n, Sustainable Use License version 1.0. https://raw.githubusercontent.com/n8n-io/n8n/master/LICENSE.md . Current file. Read at source on 2026-10-07.
6. n8n, pricing. https://n8n.io/pricing/ . Undated page. Read at source on 2026-10-07 (through an extraction tool).
7. Make and Zapier, pricing. https://www.make.com/en/pricing and https://zapier.com/pricing . Undated pages. Read at source on 2026-10-07 (through an extraction tool; Zapier’s $19.99 also seen in the raw page, Make’s raw page refused us).
8. Instantly, “Cold Email Benchmark Report 2026” and pricing. https://instantly.ai/cold-email-benchmark-report-2026 (data from 2025-01-01 to 2025-12-18) and https://instantly.ai/pricing . Read at source on 2026-10-07. A vendor’s own data.
9. Freelancer.com, open n8n jobs. https://www.freelancer.com/jobs/n8n/ . Live listing, one-day snapshot. Read at source on 2026-10-07.
10. Freelancer.com, fees and charges. https://www.freelancer.com/feesandcharges . Undated page. Read at source on 2026-10-07. Jobs and hours needed are our estimate.
11. Freelancer.com, n8n freelancers. https://www.freelancer.com/freelancers/skills/n8n . Live listing. Read at source on 2026-10-07 (through an extraction tool). Not cited in the text: the ten profiles shown first, out of 210, asked $10 to $55 an hour; they are paid placements.
12. Skool, Morningside’s free AI automation agency community. https://www.skool.com/learn-ai/about (also seen at https://www.skool.com/ai-automation-agency-hub-8466/about ). Live page. Read at source on 2026-10-07. Seller’s own figures; the per-agency and per-client sums are our estimate.
13. AAA Accelerator, sales page. https://www.aaaaccelerator.com/ . Undated page (footer 2026). Read at source on 2026-10-07. Seller’s own figures.
14. AAA Accelerator, case studies. https://www.aaaaccelerator.com/case-studies . Undated page. Read at source on 2026-10-07. The count of people and figures is ours.
15. AAA Accelerator, terms of service. https://www.aaaaccelerator.com/legal/terms . Last updated 2026-01-01. Read at source on 2026-10-07.
16. Trustpilot, AAA Accelerator, one- to three-star reviews and the company’s replies. https://www.trustpilot.com/review/aaaaccelerator.com?stars=1&stars=2&stars=3 ; single reviews at https://www.trustpilot.com/reviews/67e046c5a4f68501fc980c8f , https://www.trustpilot.com/reviews/67d1bf3392c5c89b6367b40d and https://www.trustpilot.com/reviews/6a561806a9b2875edc4719f5 . Reviews shown with dates from 2025-01-21 to 2026-07-13. Read at source on 2026-10-07 through the reader r.jina.ai; direct download was refused.
17. Trustpilot, AAA Accelerator, five-star reviews by date, pages 1 and 2. https://www.trustpilot.com/review/aaaaccelerator.com?stars=5&sort=recency and https://www.trustpilot.com/review/aaaaccelerator.com?stars=5&sort=recency&page=2 . Reviews shown with dates from 2025-10-20 to 2026-09-17. Read at source on 2026-10-07 through the reader r.jina.ai. The count of reviews that mention a client is our estimate.
18. Skool, AI Profit Boardroom. https://www.skool.com/ai-profit-lab-7462/about . Live page. Read at source on 2026-10-07. Monthly revenue is our estimate.
19. Vapi and Retell, pricing. https://vapi.ai/pricing and https://www.retellai.com/pricing . Undated pages. Read at source on 2026-10-07 (through an extraction tool). The per-minute sum and monthly cost are our estimate.
20. HighLevel, pricing. https://www.gohighlevel.com/pricing . Undated page. Read at source on 2026-10-07 (through an extraction tool).
21. OpenAI, API pricing. https://developers.openai.com/api/docs/pricing . Undated page. Read at source on 2026-10-07 (through an extraction tool).
22. US Bureau of Labor Statistics, Business Employment Dynamics, survival of establishments. https://www.bls.gov/bdm/us_age_naics_00_table7.txt . Data to March 2025. Read at source on 2026-10-07 (through an extraction tool).
23. Federal Trade Commission, complaint in FTC v. Air AI (D. Ariz.). https://www.ftc.gov/system/files/ftc_gov/pdf/Air.ai-Complaint.pdf . 2025-08-25. Read at source on 2026-10-07.
24. Federal Trade Commission, press release on the Air AI complaint. https://www.ftc.gov/news-events/news/press-releases/2025/08/ftc-sues-stop-air-ai-using-deceptive-claims-about-business-growth-earnings-potential-refund . 2025-08-25. Read at source on 2026-10-07.
25. Federal Trade Commission, press release on the Air AI settlement. https://www.ftc.gov/news-events/news/press-releases/2026/03/air-ai-its-owners-will-be-banned-marketing-business-opportunities-settle-ftc-charges-company-misled . 2026-03-24. Read at source on 2026-10-07 (through an extraction tool); the order itself was not opened.
26. Federal Trade Commission, “Operation AI Comply” press release. https://www.ftc.gov/news-events/news/press-releases/2024/09/ftc-announces-crackdown-deceptive-ai-claims-schemes . 2024-09-25. Read at source on 2026-10-07 (through an extraction tool).
27. Federal Trade Commission, business guidance on the Business Opportunity Rule. https://www.ftc.gov/business-guidance/resources/selling-work-home-or-other-business-opportunity-revised-rule-may-apply-you . November 2011. Read at source on 2026-10-07.
28. MIT NANDA, “State of AI in Business 2025”. https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf . July 2025. Secondary: the full report was read on 2026-10-07, but on a third-party copy, not MIT’s own site.
29. n8n, expert partner directory and requirements. https://experts.n8n.io/ , https://n8n.io/experts/ and https://n8n.io/expert-partners/ . Undated pages; the directory shows 48 results over two pages. Read at source on 2026-10-07.
30. n8n, customer acceptable use policy. https://n8n.io/legal/customer-acceptable-use-policy/ . March 2026. Read at source on 2026-10-07.
31. n8n, security advisories on GitHub, pages 1 and 2. https://github.com/n8n-io/n8n/security/advisories and https://github.com/n8n-io/n8n/security/advisories?page=2 . Advisories dated 2026-09-30. Read at source on 2026-10-07.
32. LinkedIn, user agreement. https://www.linkedin.com/legal/user-agreement . Effective 2025-11-03. Read at source on 2026-10-07.
33. Federal Trade Commission, “CAN-SPAM Act: a compliance guide for business”. https://www.ftc.gov/business-guidance/resources/can-spam-act-compliance-guide-business . August 2023, edited January 2024. Read at source on 2026-10-07.
34. Google, email sender guidelines. https://support.google.com/a/answer/81126?hl=en . Requirements effective 2024-02-01. Read at source on 2026-10-07.
35. Federal Communications Commission, Declaratory Ruling FCC 24-17. https://docs.fcc.gov/public/attachments/FCC-24-17A1.pdf . Released 2024-02-08. Read at source on 2026-10-07.
36. Anthropic, usage policy. https://www.anthropic.com/legal/aup . Effective 2025-09-15. Read at source on 2026-10-07.
37. OpenAI, usage policies. https://openai.com/policies/usage-policies/ . Updated 2025-10-29. Read at source on 2026-10-07 through the reader r.jina.ai; the direct page refused us.
38. Meta, terms for the WhatsApp Business Platform. https://www.facebook.com/legal/Meta-Terms-for-WhatsApp-Business-Platform . Effective 2026-09-23. Read at source on 2026-10-07 (partly as a summary).
39. EU AI Act, Article 50, unofficial copy. https://artificialintelligenceact.eu/article/50/ . Application date shown as 2026-08-02. Secondary.
40. US Internal Revenue Service, self-employment tax. https://www.irs.gov/businesses/small-businesses-self-employed/self-employment-tax-social-security-and-medicare-taxes . Last reviewed 2026-06-27. Read at source on 2026-10-07.
41. Twilio, A2P 10DLC registration for US text messages. https://www.twilio.com/docs/messaging/compliance/a2p-10dlc . Last modified 2026-07-07. Read at source on 2026-10-07. Not cited in the text; see “What we did not verify”.
42. Skool, pricing. https://www.skool.com/pricing . Undated page. Read at source on 2026-10-07.
43. Skool, affiliate programme. https://www.skool.com/affiliate-program . Undated page. Read at source on 2026-10-07.
44. Make, affiliate programme. https://www.make.com/en/affiliate . Undated page. Read at source on 2026-10-07.
45. HighLevel, affiliate programme. https://www.gohighlevel.com/affiliate . Undated page. Read at source on 2026-10-07.
46. Whop, pricing. https://whop.com/network/pricing/ . Undated page. Read at source on 2026-10-07.
47. n8n community forum, Jobs category. https://community.n8n.io/c/jobs/13 . Threads dated 2026-09-16 to 2026-10-06. Read at source on 2026-10-07. The count of hiring posts is our estimate.
48. n8n community forum, “Seeking estimates: n8n developer for a small RFQ-to-quote prototype”. https://community.n8n.io/t/seeking-estimates-n8n-developer-for-a-small-rfq-to-quote-prototype/312281 . 2026-09-10, replies to 2026-10-07. Read at source on 2026-10-07. Price statistics are our estimate.
49. n8n community forum, “Freelancer required: n8n AI automation for manufacturing”. https://community.n8n.io/t/freelancer-required-n8n-ai-automation-for-manufacturing/308576 . 2026-08-20, replies to 2026-10-07. Read at source on 2026-10-07.
50. Morningside AI. https://www.morningside.ai/ . Undated page. Read at source on 2026-10-07 (through an extraction tool).
51. n8n-MCP project on GitHub, and its author’s consultancy site. https://github.com/czlonkowski/n8n-mcp and https://www.aiadvisors.pl/en . Undated pages. Read at source on 2026-10-07.
52. Nick Saraev, “About”. https://nicksaraev.com/about/ . Undated page. Read at source on 2026-10-07. Self-reported figures.
53. Nick Saraev, “The next few months of my career”. https://nicksaraev.com/the-next-few-months-of-my-career/ . 2024-09-03. Read at source on 2026-10-07 (through an extraction tool). Not cited in the text; an earlier statement of the same author’s income.
54. Skool, Maker School. https://www.skool.com/makerschool/about . Live page. Read at source on 2026-10-07.
55. US Census Bureau, “AI use by businesses”. https://www.census.gov/library/stories/2026/05/ai-use-businesses.html . 2026-05-26. Read at source on 2026-10-07.
56. Nick Saraev, “5 more automations you can sell today for $1,500 or $10,000”. https://nicksaraev.com/5-more-automations-you-can-sell-today-for-1-500-or-10-000/ . 2024-09-01. Read at source on 2026-10-07 (through an extraction tool).
57. n8n, public template data feed, all 13,149 templates. https://api.n8n.io/api/templates/search?rows=100&page=1 (and following pages). Live data. Read at source on 2026-10-07. The counts and price statistics are our estimate from that data.
58. n8n community forum, “[HIRING] Part-time ‘household CTO’: n8n + Notion + local LLM home setup”. https://community.n8n.io/t/hiring-part-time-household-cto-n8n-notion-local-llm-home-setup-la-remote/315978 . 2026-09-23. Read at source on 2026-10-07.
59. n8n community forum, “[Hiring] Contract implementer (remote, 1099)”. https://community.n8n.io/t/313974 . 2026-09-16. Read at source on 2026-10-07.
60. n8n community forum, “[HIRING] N8N AI Automation Developer (Remote)”. https://community.n8n.io/t/hiring-n8n-ai-automation-developer-remote/298577 . 2026-06-07. Read at source on 2026-10-07.
61. n8n community forum, “[Hiring] AI and CRM Implementation Consultant + Partner Coach | Remote | $6,000/mo”. https://community.n8n.io/t/307267 . 2026-08-12. Read at source on 2026-10-07.
62. European Central Bank euro reference rate, 1 euro = $1.1269, as served by the Frankfurter rates service. https://api.frankfurter.app/2026-10-06?from=EUR&to=USD . 2026-10-06. Secondary (a service that republishes the ECB’s rates), read on 2026-10-07. The $140 total is our estimate.
63. n8n community forum, “Lead form → AI score → Slack in 72h, $249, paid after you see it run”. https://community.n8n.io/t/309728 . 2026-08-26. Read at source on 2026-10-07.
64. Upwork, fourth quarter and full year 2025 results, as filed with the US Securities and Exchange Commission. https://www.sec.gov/Archives/edgar/data/1627475/000162747526000005/exhibit991-upwork4q25andfu.htm . 2026-02-09. Read at source on 2026-10-07. The company’s own figures.
65. Upwork, first and second quarter 2026 results, as filed with the US Securities and Exchange Commission. https://www.sec.gov/Archives/edgar/data/0001627475/000162747526000033/upwork1q26-pressrelease.htm and https://www.sec.gov/Archives/edgar/data/0001627475/000162747526000046/upwork2q26-pressrelease.htm . 2026-05-07 and 2026-08-10. Read at source on 2026-10-07. The company’s own figures.
66. Zapier, Solution Partner directory. https://zapier.com/partnerdirectory . Live page showing 556 results. Read at source on 2026-10-07. The review counts by partner are our estimate from 402 of the 556 entries.
67. Zapier, partner page for AndrewJDavison. https://zapier.com/partnerdirectory/partner/andrewjdavison . Live page. Read at source on 2026-10-07 (through an extraction tool; the review count was also seen in the raw directory page).
68. NoCode MBA, “How Luhhu makes money helping other businesses use Zapier”. https://nocode.mba/interviews/how-luhhu-makes-money-helping-other-businesses-use-zapier . Undated interview (after 2019), with an update dated 2023. Read at source on 2026-10-07. Self-reported figures.
69. Companies House (UK), Luhhu Ltd, company number 11721564, filing history. https://find-and-update.company-information.service.gov.uk/company/11721564/filing-history . Accounts to 2025-12-31, filed 2026-01-22. Read at source on 2026-10-07. Unaudited accounts; Companies House does not check their accuracy.
70. n8n, expert partner pages, 45 opened. https://experts.n8n.io/partner/pulpsense (and other pages under https://experts.n8n.io/partner/ ). Live pages. Read at source on 2026-10-07. The count of lowest accepted budgets is our estimate.
71. n8n, expert partner page for AI Advisors. https://experts.n8n.io/partner/aiadvisors . Live page. Read at source on 2026-10-07.
72. Mixergy, interview with the founder of PressW, “$2 million a year AI engineering firm”. https://mixergy.com/interviews/2-million-year-ai-engineering-firm/ . 2026-01-12. Read at source on 2026-10-07. Self-reported figures; an interview sponsored by Zapier.
73. Commercial Observer, article on Rezora. https://commercialobserver.com/2026/03/leads-generator-rezora-io-cold-calls/ . 2026-03-24. Read at source on 2026-10-07. Self-reported figure.
74. Mixergy, interview with Rezora’s founders, “AI automation that makes cold calls”. https://mixergy.com/interviews/ai-automation-that-makes-cold-calls/ . 2026-01-08. Read at source on 2026-10-07 (the transcript is cut short for non-members). Self-reported figures; an interview sponsored by Zapier.
75. Zapier, partner page for Flow Digital. https://zapier.com/partnerdirectory/partner/flow-digital . Live page. Read at source on 2026-10-07.
76. TrustMRR, list of all listings and single listing pages. https://trustmrr.com/startup-sitemap.xml , https://trustmrr.com/startup/automate-chill and https://trustmrr.com/startup/reddit-agency . Live pages, 11,378 listings. Read at source on 2026-10-07. The name scan is our estimate.
77. MarketBeat, summary and transcript of Fiverr’s second quarter 2026 results. https://www.marketbeat.com/earnings/reports/2026-7-29-fiverr-international-ltd-stock/ . 2026-07-29. Secondary; Fiverr’s own release refused us.
78. betonai.net, article claiming income data from 81 automation freelancers. https://betonai.net/n8n-vs-make-vs-zapier-2026-which-automation-platform-actually-earns-freelancers-the-most-money-real-income-data-from-81-operators/ . Undated, 2026. Read at source on 2026-10-07. Cited only to explain why we do not use it.
79. Mixergy, interview with the founder of an AI consulting firm, “I make $4.5 million implementing AI”. https://mixergy.com/interviews/i-make-4-5-million-implementing-ai/ . 2026-04-23. Read at source on 2026-10-07. Self-reported figures; an interview sponsored by Zapier.
80. Zapier, “Agents of Scale” podcast, episode 8, transcript. https://agentsofscale.transistor.fm/8/transcript . 2025-09-18. Read at source on 2026-10-07. The guest’s own account on a vendor’s podcast.
81. Mixergy, interview with Zapier’s chief executive, “Building companies using Zapier’s AI automations”. https://mixergy.com/interviews/building-companies-using-zapiers-ai-automations/ . 2025-12-19. Read at source on 2026-10-07. A vendor’s own example.
82. Anonymous Reddit post by an agency owner on a year of selling automations, as copied by a third-party site. https://reddit.sentinel-team.org/posts/1rdiwhf/snapshots/2026-02-27T17%3A02%3A05.77742Z . Post from about 2026-02-25 (our working: the copy showed “Posted 224 days ago” when opened on 2026-10-07); copy dated 2026-02-27. Secondary; Reddit refused us.
83. n8n community forum statistics and the n8n code project on GitHub. https://community.n8n.io/about.json and https://api.github.com/repos/n8n-io/n8n . Live data. Read at source on 2026-10-07.
84. n8n blog, announcement of its Series C funding. https://blog.n8n.io/series-c/ . 2025-10-09. Read at source on 2026-10-07. The company’s own figures.
85. TechCrunch, “Fair code pioneer n8n raises $60M for AI-powered workflow automation”. https://techcrunch.com/2025/03/24/fair-code-pioneer-n8n-raises-60m-for-ai-powered-workflow-automation . 2025-03-24. Read at source on 2026-10-07.
86. n8n community forum, Jobs category, public data for the first page. https://community.n8n.io/c/jobs/13.json . Threads dated 2026-09-07 to 2026-10-06. Read at source on 2026-10-07. The reply statistics are our estimate.
87. n8n community forum, hiring post by a new agency. https://community.n8n.io/t/316792 . 2026-09-26. Read at source on 2026-10-07.
88. n8n community forum, two hiring posts. https://community.n8n.io/t/319256 and https://community.n8n.io/t/315672 . 2026-10-06 and 2026-09-22. Read at source on 2026-10-07.
89. DigitalOcean, server (“Droplet”) pricing. https://www.digitalocean.com/pricing/droplets . Undated page. Read at source on 2026-10-07.
90. US District Court, District of Arizona, FTC v. Air AI, Amended Stipulated Order, docket entry 46, on CourtListener. https://storage.courtlistener.com/recap/gov.uscourts.azd.1455303/gov.uscourts.azd.1455303.46.0.pdf . Signed 2026-03-30, filed 2026-03-31. Read at source on 2026-10-07.
91. Apify blog, profile of Nick Saraev. https://blog.apify.com/nick-saraev-and-apify/ . 2025-02-25. Read at source on 2026-10-07. Written by a company that pays him for referrals; self-reported figures.

Corrections

If you are quoted or named on this page and think something is wrong, want your reply shown beside it, or want to be removed, write to [email protected]. We aim to reply within 14 days, and always within one month. Factual errors are corrected with a dated note here. A sentence that is seriously disputed comes down while we check it.

Change log:

• 2026-10-07: Replaced “When it can work” with “The upside” and “What it takes to compete”, using a second research pass with web search. Changed the label from “Hard without contacts” to “Builders with industry contacts”, because the accounts found say contacts can be built while doing the work. Corrected two statements on the Air AI case: the court has approved the settlement⁹⁰, and the $250,000 loss figure appears only in the FTC’s announcement, not in the complaint²³²⁴. Added a server price, which the page had said it did not source⁸⁹. Recounted the n8n partner budget bands: 43 of 45 pages show them, and 42 of the 43 firms say they accept nothing under $1,000 (the first count had 29 of 30)⁷⁰.
• 2026-10-07: wording about named accounts and companies reviewed; two example tweets from small accounts were replaced with tweets from larger ones, one reviewer’s allegation and a company’s balance-sheet figures were removed, and several of our inferences are now marked as such.