Most articles about making money with artificial intelligence pick a number that sounds impressive and work backwards from it. A million. Ten thousand a day. The number is chosen for the headline, not because anything in the data supports it.
A hundred thousand dollars in a year is a different case, and it is worth stating why before anything else: it is the first target on that list where the published trajectories actually reach the number. Not a stretch, not an outlier. The ordinary documented path lands there.
This is what that path looks like, what it costs to start, and what is genuinely selling right now.
The number, and why it is the right one
A hundred thousand a year averages 8,333 a month. Nobody starts there, so the shape that matters is the ramp.
Published year-one timelines for solo operators working full time on AI service work run roughly like this: months one to three produce between zero and five thousand in total while the first client is found, months four to six settle at three to ten thousand a month with two or three clients, and months seven to twelve reach eight to twenty-five thousand a month once the offer is productised and outreach is consistent.
Take the conservative middle of each band - two thousand across the first quarter, six thousand a month in the second, twelve thousand a month in the third and fourth - and the year totals about ninety-two thousand.
That is the point. A hundred thousand is not a stretch goal on this path. It is roughly where the ordinary version of it arrives, and the difference between ninety-two and a hundred is two extra clients or one price increase.
Compare that with a million in the same period, which requires a peak of a hundred and sixty-seven thousand a month and about twenty-one clients simultaneously. That is a different sport. A hundred thousand needs, at the peak, five clients at twenty-five hundred a month. Five relationships. You can hold five relationships in your head.
What it costs to start
This is the part that surprises people, so here are current prices.
ChatGPT Plus is twenty dollars a month. Claude Pro is twenty. Perplexity Pro is twenty. Descript, for video with automatic transcription and filler-word removal, is twenty-four. Canva Pro is eighteen a month or a hundred and forty-four a year. A typical working stack in 2026 runs fifty to eighty dollars a month.
Call it nine hundred and sixty dollars for the year. Against a hundred thousand target that is one per cent.
The barrier to entry is not money and it is not the tools. It has not been for two years. Which means the barrier is somewhere else, and the rest of this is about where.
What is actually selling
Demand for AI-related freelance skills on Upwork grew 109% year over year as of February 2026. Within that, AI video generation and editing grew 329% and is the fastest-moving category on the platform.
The specific services with live demand right now:
**Short-form video editing and AI voiceover.** The fastest-growing category, driven by brands that need dozens of ad variants a month because performance creative decays with repetition.
**Chatbot and agent setup.** Businesses are paying between three hundred and fifteen hundred dollars for a working 24/7 support agent built on Voiceflow, Botpress or Stack AI. This is the single most accessible entry point on the list: the platforms are visual, the deliverable is obvious, and the buyer can see it working in a demo.
**Product-image editing for e-commerce.** Catalogue imagery at a fraction of studio cost and turnaround.
**SEO content packages, UX microcopy, data cleaning and analysis.** Less glamorous, steady demand.
**Ghostwriting, prompt engineering and UGC video** are reported as the highest-margin categories, because they reward judgement rather than production polish.
Note what changed. Knowing how to prompt a model is now the baseline expectation, not the service. Clients are hiring people who build production systems, integrate them into existing workflows and can point at a measurable business outcome. "I know ChatGPT" is not an offer any more. "Your support queue drops 40% and here is the dashboard" is.
The niches where this works
The businesses worth approaching share three properties: they have repetitive processes that touch revenue directly, they are not technical enough to build the automation themselves, and they can sustain a thousand dollars a month or more in fees.
Four that are documented as working in 2026:
Marketing agencies, for client reporting pipelines. They produce the same report for every client every month and hate it.
Real estate teams, for lead routing and follow-up sequences. A lead that sits for an hour is often a lost lead, and this is measurable.
SaaS companies between fifty and five hundred employees, for trial nurture and churn detection. Large enough to have the problem, small enough not to have a data team.
E-commerce brands above a million in revenue, for order operations and support triage.
What none of these want is a chatbot in the abstract. They want a specific recurring annoyance to stop.
The twelve months, in order
**Months one to three. Pick one service and one industry, then build the thing before you have a client.** Not a portfolio of screenshots - a working demo using a real company's public data. Somebody's actual product catalogue, somebody's actual FAQ page. Published outreach data says a focused operator sending fifteen to twenty-five targeted messages a day lands a first discovery call within one to two weeks and a paying client within four to eight. Expect to earn very little in this quarter. Its job is the first case study.
**Months four to six. Convert the first client into a repeatable offer.** This is where most people stall, because the second sale feels like the first one all over again. It should not: you now have a named business you helped and a number you moved. Two or three clients at this stage puts you between three and ten thousand a month.
**Months seven to twelve. Raise prices and stop taking one-off work.** Everything you sell should be monthly. A one-off delivery leaves you back at zero next month; a retainer means January's revenue exists before January starts. This is where the eight to twenty-five thousand a month band comes from, and where the year is decided.
Six things that compress the timeline
**Do the work before the pitch.** Take a target company's real product page, real catalogue, real support FAQ, and build the thing. Then send it with no pitch attached. This converts at a rate no cold email approaches, because you have removed every step between them and seeing the result.
**Price against their current invoice, not your cost.** Your cost is nearly zero and falling, which is precisely why it is irrelevant. The relevant number is what they pay today - the agency retainer, the freelancer, the salary of whoever does it now.
**Sell the outcome, never the model.** The moment your pitch names a model, you are competing with everyone who can name the same model, and the buyer starts wondering whether their intern could do it. Nobody has a budget line for AI. They have budget lines for support, for content, for advertising production.
**Go monthly from the first contract.** "Two hundred images" is a project. "Two hundred images a month" is a business. The client often prefers it, because it removes their own repeat decision.
**Pick the boring niche.** Marketing agency reporting is dull and pays. Anything that demos well is crowded.
**Keep the failure knowledge.** After twenty deliveries you will know where the tools break on real data - which products confuse the image model, which accents the voice model mangles, which support questions need a human. That knowledge, not the subscription, is what stops the client doing it themselves.
What actually kills it
The documented failure pattern is specific and worth memorising, because it is what most people do.
Someone picks three services rather than one, builds half of each, sends no consistent outreach, finds no client within thirty days and concludes that AI work does not pay. The tools were never the problem. Thirty days is simply less than the four to eight weeks the published data says a first client takes, and three half-built offers is not the same as one finished one.
The second killer is one-off work. A year of project fees can total a respectable number and leave you starting from zero every January.
The third is competing on price in a category where the price is heading to zero. If your only advantage is being cheaper, your margin has an expiry date visible from here.
The honest summary
A hundred thousand dollars in a year from AI service work is available to a person with no team, no funding and a thousand dollars of annual software cost. That is a real and unusual thing, and the published trajectories support it.
What it requires is not technical. It is one service instead of three, one industry instead of any, consistent outreach for longer than thirty days, and the discipline to sell monthly rather than once. Every one of those is a decision rather than a skill, which is why the path is genuinely open to most people and why most people will not take it.
What we do
We build AI agents and the trading terminals they run in, and we publish the scored record rather than the pitch. Every forecast is serialised, hashed and timestamped into a Bitcoin block before publication, then scored in public with the misses on the same page as the hits. When our stated 50% intervals turned out to be containing 84% of outcomes, that record is where we found it, and we published the diagnosis before we had a correction.
We report no income from the path described above, because we do not run it. Everything here is sourced from published 2026 platform data and stated as such, which is the standard we would want applied to anyone telling us how to make money.
Educational content - not financial advice.