YouMind
Войти

$820 A MONTH ON 4 AI SUBSCRIPTIONS. I CANCELLED 3, KEPT ONE FOR CODE, AND STOPPED PICKING MODELS

@starmexxx
АНГЛИЙСКИЙ17 нояб. 2026 г.
183K
22
1
6
22

Суть

The author details replacing four expensive AI subscriptions with a single AI agent named Viktor to automate daily income stream research, saving $620 monthly while maintaining high-quality output through smart model routing.

I paid for 4 AI subscriptions to find income streams and lost hours picking which to open.

Follow me. I’m

@starmexxx , I track how AI tools are creating new income streams most people haven’t heard of yet.

New models ship every week, and every launch thread promises the same thing: this one reasons better, that one is cheaper, the next one is faster. For a year I kept paying for all of them, because finding new ways to make money with AI is literally my job, and I didn’t want to miss the one model that sees something first.

By November that habit cost me $820 a month and a big part of every morning. Not on the actual work, but on deciding which tab to open and explaining the same context again and again to each model.

So I moved the whole hunt into one Slack channel and handed it to Viktor.

Below is the full setup: what I was paying for, the exact goal I gave Viktor, how a cheap model and a top model split the work between them, what came back on day one, where every single credit went, and the final math. Every prompt in this article is copy-paste ready.

1/

What I was actually paying for

text
1SUBSCRIPTION PLAN PER MONTH WHAT I USED IT FOR
2ChatGPT Pro $200 research, quick lookups
3Claude Max 20x $200 code, hard reasoning, writing
4Grok SuperGrok Heavy $300 what's moving on X right now
5Midjourney Mega $120 visuals for posts
6-------------------------------------------------------------------------
7TOTAL 4 subscriptions $820 / month

None of these is a bad product, and that’s exactly the trap. Each one is the best at something, so you end up paying for all four and using each of them for twenty minutes a day.

Four tools meant four tabs, four separate chat histories and four places where my notes lived. A promising income stream I found in the morning would end up split across three chats by lunch, and half the source links would get lost somewhere in between.

starmex - inline image

2/

The real cost was explaining everything three times

The $820 was the visible cost. The hidden one was context.

Every model starts from zero. So a normal morning looked like this:

text
1STEP 1 research tab: scan X and Reddit for new AI money angles
2STEP 2 second tab: paste findings, explain what I'm looking for, ask for a summary
3STEP 3 summary too shallow → third tab, explain the whole context again
4STEP 4 want a visual for the post → fourth tab, explain it one more time
5STEP 5 copy the result into notes, lose half the source links on the way
6STEP 6 finally start the actual work

By the time I got to step 6, I had described the same goal to three different models, and none of them remembered anything from yesterday.

That’s what I actually wanted to get rid of. Not the models, but the constant switching and re-explaining between them.

starmex - inline image

3/

Hiring Viktor instead of opening another tab

Viktor is an AI employee who lives in Slack and Microsoft Teams. You don’t open him in a new tab and you don’t start every conversation from scratch: you message him like a teammate, give him a goal once, and he keeps running it on a schedule in its own channel.

For context on scale, 79,000+ teams have hired Viktor so far, and he connects to 9,800+ tools.

The setup took me about five minutes:

text
11. viktor.com → sign up
22. connect your Slack workspace (or Microsoft Teams)
33. open Viktor in Slack → "New goal"
44. describe the goal in plain English
55. team mode: leave it on Smart
66. confirm the schedule when Viktor asks

Tip: if your team doesn’t use Slack yet, create a free workspace just for Viktor. One channel per goal keeps every result, every source and every decision in one readable place.

4/

The exact goal I gave Viktor

This is the full prompt. Copy it as it is, or swap the sources for your own niche:

markdown
1Find new ways to make money with AI before they go mainstream.
2
3Every morning, scan X, Reddit and Product Hunt for new AI income
4streams and post a short list here, using the Cheap model.
5
6Every Friday, write a deeper report on the 3 best ones with how
7to start, using the top model.

Viktor didn’t start the schedule blind. First he came back with a plan:

text
1JOB WHEN WHAT MODEL
2Daily 08:00 UTC short list of new AI income streams Cheap
3Weekly Fri 09:00 UTC deep report on the week's top 3 top

He even set the Friday report one hour after the morning list on purpose, so that the weekly report can include that day’s findings too.

Then he asked one question: “Should I go with that?” Nothing ran on a schedule until I clicked yes.

5/

Why the work is split between two models

This is the core of the whole setup. Not every job needs the most expensive model, and paying top-model prices for a daily scroll through Reddit is exactly how AI bills get out of hand.

text
1JOB HOW OFTEN WHAT IT NEEDS MODEL
2Daily scan every day volume, speed, source links Cheap
3Friday deep report once a week judgment, ranking, depth top
4Everything else on demand depends on the task Smart (team default)

The daily scan reads a lot and decides very little: find new things, keep the links, write one line per idea. A cheap model handles that well.

The Friday report is a different kind of work. It has to compare ideas, rank them for my situation and explain how to start each one. That’s where a top model actually earns its price.

Our team mode stays on Smart by default, and I only pin a specific model to a job when I know why that job needs it.

Tip: start with Smart for everything. Pin a model to a job only after you’ve seen what that job costs. Section 8 shows exactly how to check that.

starmex - inline image

6/

What came back on day one

Viktor didn’t wait until the next morning. He ran the first scan right away and posted 6 early AI income ideas, each with a source link and a rough effort level:

text
1# IDEA WHAT THE SOURCE SAYS EFFORT
21 Paid X posts through an agent marketplace creators keep 80%, Transformer.gg low
3 launched on 15 Nov
42 Selling expert recordings and judgments paid per accepted piece of work, no build
5 to AI labs median ~$80/hr on one data broker
63 Charging AI agents per API call (x402) most sellers report almost no very early
7 revenue so far
84 Niche directory built by a bot that builder says it made $217 in its low
9 mines complaints third month
105 Live translation app for churches $5/hr per language or $99 to $499 medium
11 per month
126 Agent-readiness audits for small about $500 per audit medium
13 business sites

Two things here impressed me more than the list itself.

First, Viktor flagged on his own that the income figures come from the sources and that he couldn’t verify them. For the church translation app he even pointed out that the $14K a month figure came from a newsletter and not from the founder. That’s the kind of note I’d expect from a careful analyst, not from a scraper.

Second, he was upfront about his method: X posts are only visible when logged in, so he covered X through web search, while Reddit and Product Hunt he read directly.

He also picked one to test first: #2, because it has a published rate and needs nothing built.

starmex - inline image

7/

The Friday report, run on day one

I didn’t want to wait until Friday, so I asked Viktor to run the weekly report right away on the first list. The top model narrowed it down to 3 ideas and ranked them for my exact situation. Two of the three that made the cut:

text
1IDEA THE NUMBERS THE CATCH
2Agent-readiness audits $400 audit, credited against a it's getting crowded
3for local businesses $600 fix sprint
4Paid X posts on Transformer.gg you keep 80% of each paid post only works with
5 1,000+ followers

Then Viktor gave a direct read for an AI creator on X: paid posts fit best, because reach is exactly what AI startups pay for, and the platform’s brands are AI startups. He also named the downside without me asking: every post carries a “Paid partnership” label, which can cost you trust with your audience. Agent-readiness audits came in as a close second.

He finished with two questions that would change the ranking: my professional field and how many followers I have on X, because the field sets the expert rate.

That’s what I want from the expensive model. Not more text, but better judgment about my specific situation.

8/

The checkup: where did my credits go?

markdown
1where are my credits going?

He broke down every job that ran that day:

text
1WHAT RAN MODEL CREDITS
2Top 3 deep report top 861
3Chat: setup, price table, coordinating Smart 792
4AI creator on X ranking top 214
5Goal background research Smart 203
6First daily scan Cheap 117
7----------------------------------------------------------------
8TOTAL AT THE CHECKUP 2,187
starmex - inline image

Look at the gap between the first and the last line. The daily scan on the cheap model cost 117 credits. The deep report on the top model cost 861. That’s more than 7x for one job, and it’s exactly why running everything on one model makes no sense.

Viktor also pointed out that every web search costs 5 credits, which explains why research-heavy jobs add up faster than you’d expect.

Then he projected the month from that day’s runs:

text
1Daily scans ~3,500 credits / month
2Friday reports ~3,700 credits / month
3---------------------------------------------
4Total ~7,000 credits / month, before any chats

He said clearly that this is his estimate from day one, not a measured rate.

And then came the most useful part: one fix he proposed without me asking for it.

text
1PROPOSAL run the Friday report on the mid-tier model
2EFFECT roughly half the credits
3TRADEOFF the analysis gets a bit less deep

I approved it. Nothing changed until I clicked the button.

starmex - inline image
starmex - inline image

9/

What I cancelled and what I kept

After the first day of running the hunt through Viktor, I went through the four subscriptions one by one:

text
1SUBSCRIPTION PER MONTH DECISION WHY
2ChatGPT Pro $200 cancelled research now runs as Viktor's daily scan
3SuperGrok Heavy $300 cancelled X trends now come through the same scan
4Midjourney Mega $120 cancelled not worth $120 for occasional visuals
5Claude Max 20x $200 kept still my main tool for code
6-------------------------------------------------------------------------------
7CANCELLED $620 / month

Claude stays, and I’ll be honest about why: for code and long, hard reasoning sessions I still want direct control over the model and pick it by hand.

text
1JOB HOW I RUN IT NOW
2Code, long reverse engineering sessions Claude, picked by hand
3The daily hunt for new income streams Viktor, Cheap model
4The weekly deep report Viktor, mid-tier after the checkup
5Anything else that repeats on a schedule Viktor, I don't think about it

For one-off deep work, picking the model myself still makes sense. For anything that runs daily or weekly, picking by hand was pure waste.

10/

The math

text
1BEFORE 4 AI subscriptions $820 / month
2 the same context explained to 3-4 models every morning
3
4NOW Claude for code $200 / month
5 Viktor in one Slack channel 2,187 credits on day one
6 ~7,000 credits / month projected
7 all on reward credits so far
8 one goal, two scheduled jobs, one place for every result
9
10CHANGE 3 subscriptions cancelled $620 / month
11 1 fix approved after the checkup Friday report moved to mid-tier

11/

The whole setup in one block

text
1STEP 1 viktor.com → sign up → connect Slack or Teams
2STEP 2 New goal → paste the prompt from section 4
3STEP 3 team mode: Smart by default
4STEP 4 daily scan → Cheap model
5STEP 5 Friday report → top model
6STEP 6 confirm the schedule when Viktor asks
7STEP 7 after day one, ask: "where are my credits going?"
8STEP 8 approve only the fixes that make sense for you
9STEP 9 after week one, go through your subscriptions and cancel
10 the ones the scheduled jobs already cover
11
12RULE pin a model to a job only after you've seen what that job costs

Before you go

A year ago an “AI stack” meant paying for every new model and switching tabs all day, re-explaining the same goal to each one of them.

Now the expensive model only touches the work that actually needs judgment, the cheap one handles the volume, and I can see exactly where every credit goes before anything changes. The only subscription I kept is the one I use for code every day.

The real edge isn’t knowing which model to pick. It’s not having to pick at all.

Hire

@viktor_com for your team, free to start:

https://ref.viktor.com/starmex-x Code: STARMEX100

// The window is open. Follow @starmexxx. I’ll keep finding them before they close //

#AIemployee

Paid Partnership

Переделать в YouMind

Превратите одну вирусную статью в полноценный рабочий процесс создания контента

Собирайте источники, расшифровывайте паттерны, создавайте активы, пишите черновики и публикуйте контент из одного рабочего пространства ИИ.

Исследовать YouMind
Для авторов

Превратите ваш Markdown в аккуратную статью для 𝕏

Когда вы публикуете длинные тексты, изображения, таблицы и блоки кода, форматирование в 𝕏 становится мучением. YouMind превращает полный черновик в Markdown в чистую статью, готовую к публикации в 𝕏.

Попробовать Markdown для 𝕏

Другие паттерны для анализа

Недавние виральные статьи

Смотреть другие виральные статьи