Ham

Writing about AI, automation, and practical use cases. No hype, just what actually works.

[Ham @0xHam44]

Everyone is talking about the 97 GB he found. Nobody is talking about the 7 words he typed.

He typed "my mac is full, find out why." Claude mapped the disk, sorted the junk and built the cleanup UI on its own.

Now picture that running every Monday on every laptop in a 20 person design studio.

Not a cleaner app. Not a sysadmin. A weekly agent that asks before deleting anything.

That's not a disk tool. It's an IT department that costs less than one Adobe seat.

$12 a month in tokens. $400 a month per office.

15 clients. $6,000/month. 97% margin.

No installs. No tickets. No consultants.

People are still paying $29 for the app.
A few are already charging $400 for the prompt.

[Jay Patel @jayypatel18]

opus 5.5 is unreal

asked it why my mac was full. it mapped my whole disk, found 97gb of junk and built this ui

people are paying $29 for this 💀

[Ham @0xHam44]

@elonmusk Sub second replies on customer calls is wild

[Ham @0xHam44]

@zefirium didnt know long chats re-read 80-90K per step, new chat per task from now on

[Ham @0xHam44]

Do you actually understand what's happening.

The company with the most aggressive model lab on X just put a rival's brain inside its own agent. That same setup now runs the back office of 9 dental clinics for one guy in Kraków. $1,800 a month. Per clinic.

Not a chatbot. Not a Zapier flow. A crew.

Five bots. A Chief that reads every inbound email. One for insurance paperwork. One that chases unpaid invoices. One that books and reschedules. One that writes the weekly report the owner actually reads.

Watch the group chat. The booking bot catches a cancelled slot, hands it to the outreach bot, and the slot is refilled from the waitlist in four minutes. Nobody asked it to.

That's not automation. It's a staff.

No receptionist overtime. No missed recalls. No invoices aging past 60 days.

The rule that makes it sellable: every bot drafts, files and reconciles alone. Anything a patient sees, anything that moves money, gets parked for a human. 140 drafts queued last week. 0 sent without a click.

Setup is $2,500 once per clinic. Tooling runs about $90 a month per clinic. He charges $1,800.

9 clinics. $16,200/month. 95% margin.

The pitch to the owner is one sentence: your front desk stops drowning, and nothing leaves the building without your yes.

He opened the group chat twice this week.

The clinics hadn't noticed he was gone.

[Morlex @0xMorlex]

Grok Bot just got a serious brain upgrade

With Claude Opus 5.5 entering the stack, one Bot could run an entire AI team that researches, writes code, uses tools, verifies results, and delivers finished work

Here's how to build it in 10 steps: https://t.co/SysbtXuqoJ

[Ham @0xHam44]

Do you actually understand what's happening.

A guy with zero design skills is selling spreadsheets to plumbers. $7,800 last month. He has never fixed a pipe.

That's not a template shop. It's a research machine that happens to sell files.

Every morning a Claude agent scans 60 Etsy and Gumroad shops. It hunts for one listing pulling 20x more reviews than the rest of its shop. Then it strips out who the product was for and keeps only what it does.

A wedding budget tracker becomes a job quote calculator for roofers. A freelance rate sheet becomes a markup calculator for electricians. Same math. A buyer nobody built for.

Not printables. Not planners. Calculators for people who send quotes.

Claude writes the formulas, the tabs, the example rows. The Sheet gets duplicated, locked and listed on Gumroad at $29 before lunch. Make publishes it, then pulls the sales back into the research sheet 14 days later.

No design. No inventory. No shipping.

The bill is $29 a month. Claude and Make. Gumroad takes its cut only when something sells.

41 calculators live. 9 of them carry 80% of the revenue.
$29 each. 270 sales a month. 91% margin.

The trades are the part nobody is watching. Contractors pay for anything that makes a quote faster. They don't browse for cute fonts. They search "roofing estimate spreadsheet" and buy the first one that works.

His whole pitch fits in one line. Send better quotes in 4 minutes.

Most sellers are still fighting over brides.
He found the people with truck payments.

[Ham @0xHam44]

https://t.co/J6C1rdmrse

[Ham @0xHam44]

Do you actually understand what's happening.

Most companies running an AI support bot are paying for the same answer 40 times a day.

This guy noticed. Now he charges them to stop.

Watch the left side of that screen. Three ways to ask one question. Three different hashes. Three misses. The company pays for all three.

Now watch the right side. The same three questions land inside one circle. One answer generated. Two served from memory in 40 milliseconds.

That's not a cache. It's a refund machine.

He runs it locally. An embedding model on a $600 mini PC, a numpy matrix, about 60 lines of Python. No vector database. No cloud. No new vendor.

He plugs it in front of a client's existing support bot and lets it run for two weeks. Then he shows them one number.

A Shopify store doing 9,000 support chats a month was spending $1,400 on tokens. After the cache: $520. Hit rate 61%. Replies went from 2.3 seconds to under 100ms.

He takes 30% of what he saves them. That one store pays him $264 a month.

Setup takes an afternoon. Hardware is $600 once. Running cost is electricity.

12 stores. $3,168/month. 90% margin.

The pitch is one sentence: "You keep paying for answers you already have. Let me show you how many."

Ecommerce brands, SaaS help desks, telecoms, clinics with booking bots. Anyone whose customers ask the same thing in different words.

Not a bigger model. Not another agent. Just memory.

Most teams are still paying per question.
He gets paid for the answers they never needed to generate.

[Ham @0xHam44]

https://t.co/9pXBwZbfEv

[Ham @0xHam44]

@ericzakariasson @bot Google slides export was the one i was waiting on, thx eric

[Ham @0xHam44]

@PawelHuryn Mistral large 4 tying a 27b you can run locally at 15 is rough

[Ham @0xHam44]

@LightningMike23 @bot The email scam setup cary posted is lowkey the best use case in here so far

[Ham @0xHam44]

@vavanessadev Appreciate u splitting it by researchers vs workers, sending this to my cofounder

[Ham @0xHam44]

@Michael_Fenech_ Saving the 6 rules before my overnight run tonight, appreciate it

[Ham @0xHam44]

@PawelHuryn 8 of 43 for OpenAI is rough ngl, gonna test Jev on my intake forms

[Ham @0xHam44]

@distortgeekin 2,500 PRs a month and reading them after they land is wild

[Ham @0xHam44]

https://t.co/9pXBwZbfEv

[Ham @0xHam44]

Everyone is talking about Gemini Notebook as a study tool. Nobody is talking about it as a business.

This guy builds one notebook per client and charges $1,500 to set it up. Then $400 a month to keep it fed.

His clients are small law firms. 4 to 12 lawyers. Thousands of pages of contracts, case notes and memos nobody can find when it matters.

Not a SaaS. Not an agency. One person with a Google account.

He doesn't write code. He curates sources. 40 signed contracts, the firm's templates, two years of client emails exported as PDFs, the local regulations. Nothing else goes in.

Then he writes the chat instructions. Act as a skeptical senior associate. Quote the clause and page every time. Never answer outside the sources.

A junior lawyer types "which of our NDAs auto renew with no liability cap" and gets 9 contracts with page numbers in 20 seconds.

That's not a chatbot. It's the firm's memory, with citations.

Setup takes him about 6 hours. Upkeep is 2 hours a month, adding new files and pruning old ones. The firm already pays for Google Workspace.

8 firms. $3,200/month recurring. Close to 90% margin.

No servers. No training data. No invented case law.

His pitch is one sentence: your associates lose 5 hours a week searching, he gives them back 4.

Most people are still asking it for summaries.

A few are already invoicing for it.

[Ham @0xHam44]

https://t.co/IJKKg7rkMd

[Ham @0xHam44]

@sanja_ratwo @kara_trade The fix is boring: read only on the calendar, bookings only through one approved action

[Ham @0xHam44]

Everyone is talking about AI agents. Nobody is talking about the boring ones that refuse to do things.

This guy builds one bot per client. Each one does exactly one job, and he bills $900 a month for it.

Watch the diagram. Memories, Skills, Routines, Integrations. Four boxes. That is the whole product.

His first client was a dental clinic with 14 unconfirmed appointments every Monday. The bot reads the calendar at , drafts the reminder texts, and waits. Nothing goes out until the front desk taps approve.

That's not a chatbot. It's a receptionist that knows its limits.

Most of the memory file is a list of what it will never do. Never book. Never cancel. Never quote a price. Never invent a time slot.

Clinics trust it because of those four lines, not because of the model.

Setup takes him about 6 hours. Every client runs on one shared $40/month cloud computer. API costs land around $25 per clinic.

So each clinic pays $900 and costs him roughly $30. He copies the template, swaps the memory file, connects the next calendar.

11 clinics. $9,900/month. 97% margin.

No sales team. No custom code. No support tickets at 2am.

Not a startup. Not an agency. One template, cloned eleven times.

He checked the dashboard once this week.
Eleven waiting rooms ran themselves.

[0xRafy @0xRafy]

https://t.co/Jf2Ly0YBgV

[Ham @0xHam44]

@elonmusk @grok @bot The company card on day one line got me lol
Learns your needs is the part im watching
Ty, sending this hiring guide to my cofounder

[Ham @0xHam44]

@sohmehta 30 days of replying under every niche post is literally my plan rn, ty for the push
Dming ppl who already complained about the exact problem is so underrated, way warmer than any cold list
Fixing the first 20 users setup yourself doesnt scale and thats exactly why it works