While You Panic About AI Taking Jobs, I Built $200/Mo Tools
I eliminated $200/month in AI subscriptions by building Claude Code skills. Same AI, zero recurring cost. Here's what I built and how you can too.
I was about to click “Subscribe: $29/month” on yet another AI content tool when I stopped.
Not because $29 was a lot. I’ve subscribed to worse. I have a graveyard of forgotten SaaS products auto-renewing somewhere in my credit card statement, silently draining money for tools I used exactly twice.
No, I stopped because I realized something embarrassing.
This tool was literally just a pretty wrapper around Claude. Same AI I already pay for. Same capabilities. They’d added a nice UI, a payment form, and approximately zero additional value.
I was about to pay $29/month to rent something I could build in an afternoon.
So I closed the tab. Opened Claude Code. And two hours later, I had my own version. No usage limits. No subscription. Mine forever.
That was six months ago. Since then, I’ve built six tools. I’ve eliminated $200/month in subscriptions. And I’ve realized something that changed how I think about this whole “AI is coming for your job” panic.
Hey, I’m Lakshmi: I help developers build, deploy, and distribute their SaaS without hiring a team. I also run Stacksweller and Supabyoi.
New here? Start with Why Your AI Wakes Up Every Morning With No Memory or Clean Code Is Dead.
Everyone’s Worried About the Wrong Thing
My LinkedIn feed is a horror show right now. Every other post is either “AI will take your job” or “Here’s how to survive the AI apocalypse” or some variation of “the robots are coming, repent.”
The advice is always the same. Learn to prompt. Adapt or die. Get your finances in order because disruption is coming.
Maybe it is. I don’t know the future any better than you do.
But here’s what I do know: while everyone’s stockpiling survival advice, I’ve been using those same AI tools to eliminate $200/month in software subscriptions. Tools I was paying for six months ago? I own them now. Forever. Zero recurring cost.
Same technology. Completely different mindset.
Let me show you what I mean.
The Batch PDF Processor I Built Instead of Uploading 50 Files
I had 50 research papers to process. Extract abstracts, pull out key findings, grab any data tables, save everything as searchable markdown.
Sure, Claude can read a PDF. One at a time. Upload, wait, copy the output, upload the next one, repeat 49 more times.
Old me would’ve done exactly that. Or Googled “batch PDF extraction tool,” found something that charges per page, done the math, decided it wasn’t worth it, and then manually uploaded 50 files anyway.
New me? I built a script.
Ninety minutes later, I had a skill that loops through a folder, extracts text and tables from each PDF using PyMuPDF, summarizes each one, and saves structured markdown files. Point it at a folder, go make coffee, come back to 50 organized summaries.
> Process all PDFs in ~/research/papers
[loops through 50 files]
[extracts + summarizes each]
[saves to ~/research/summaries/]No uploading files one by one. No copy-paste marathon. No usage limits. Runs locally, so nothing leaves my machine.
The tool isn’t “PDF extraction”: Claude already does that. The tool is automation. Batch processing. The boring plumbing that turns a manual 3-hour task into a 5-minute one.
The Video Transcription Pipeline That Changed How I Learn
I’m an infoproduct junkie. Courses, masterclasses, workshops: if someone’s selling knowledge in video form, I’ve probably bought it. My Teachable and Gumroad purchase history is embarrassing. Hours and hours of content sitting in various dashboards, waiting to be watched.
Old me would take notes while watching. Pause, scribble, play, pause, scribble. Retain maybe 30% of it. Forget the rest within a week.
New me? I feed the videos to my video-distill skill. It transcribes everything with Whisper, distills it into readable chapters with Claude, and exports to EPUB.
Two hours of course video becomes a 12,000-word mini-book on my Kindle that I can search, highlight, and reference forever.
Build time: 2 hours.
Previous cost: $50-200 per course for transcription services (or just... not having transcripts and forgetting everything).
The ROI on courses went from “eh, probably worth it” to “this is a no-brainer.”
The Pattern Nobody Wants to Admit
Here’s what I noticed while building these:
Every AI SaaS is a thin wrapper around the same AI you already have access to.
Batch file processing? A loop + Claude.
Video transcription? Whisper + Claude.
Content research? Web fetch + Claude.
Cross-posting? Template formatting + Claude.
Writing assistant? Prompt engineering + Claude.
The “product” is convenience packaging. A nice UI, hosting, a payment gateway, customer support. Sometimes that’s worth paying for.
But most of the time? You can build 80% of what you need in under two hours.
I’ve done this six times now. Total build time: about 14 hours. One weekend, spread across a few months.
Total monthly savings: $199.
Total annual savings: $2,388.
Tools I now own forever: 6.
The Real Question Nobody’s Asking
While everyone argues about whether AI will take their job, here’s what I’m thinking:
It’s not AI vs humans. It’s humans with AI vs humans without.
The lawyer who refuses to use AI for contract review loses to the lawyer who uses it and handles 5x more clients.
The developer who doesn’t use AI for code generation loses to the developer who does and ships features in days instead of weeks.
The writer who thinks AI is “cheating” loses to the writer who uses it for research and drafts 10x more content.
The people who lose their jobs to AI won’t be the ones whose work AI can theoretically do.
They’ll be the ones who didn’t use AI and got outpaced by someone who did.
What I Actually Do Instead of Panicking
I audit my subscriptions. Every few months, I go through my recurring charges and ask: “Is this just an AI wrapper?”
If yes, I build a replacement. If the replacement covers 80% of my use cases, I cancel the subscription.
Here’s my current hit list:
Batch PDF processing: replaced manual uploads with pdf-reader skill (90 min)
Video transcription: replaced $50-200/course services with video-distill skill (2 hours)
Ebook formatting: replaced $30/book services with epub-builder skill (1 hour)
Content research: replaced $29/mo tools with compose skill (3 hours)
Cross-posting: replaced $50/mo tools with distribute skill (2 hours)
Writing assistant: replaced $20/mo Sudowrite etc with fiction-writer skill (4 hours)
Not everything is worth building. QuickBooks? Keep paying. Complex Zapier automation with 50 integrations? Probably keep paying. Hosted databases? Definitely keep paying.
But simple AI wrappers that just call Claude with a prompt and charge you monthly for it? Those are dying. You can own that.
This Is What Leverage Actually Looks Like
When you own your tools, you can modify them to fit your exact workflow. Combine them in ways SaaS products can’t. Build competitive advantages nobody else has.
My distribute skill cross-posts to five platforms in one command. Most people manually post to each: different formatting, different copy, different everything. That’s 30 minutes per post. I do it in 2 minutes.
Over a year, that’s 50+ hours saved. For one skill.
My video-distill skill turns courses into searchable mini-books. Most people watch courses once and forget 80% of it. I have permanent reference material I can search and review anytime.
This is how one-person operations beat ten-person teams. Not by working harder. By owning tools that make you 10x faster.
Two Paths
The AI revolution is here. Not coming. Here.
Path A: Read about how AI is going to disrupt your career. Worry about it. Prepare for the worst. Maybe it happens, maybe it doesn’t. Either way, you spent months anxious instead of building.
Path B: Use AI to eliminate expenses, build tools, ship faster, own your stack. If disruption comes, you’re already leveraged. If it doesn’t, you still saved $2,400/year and got 10x faster.
I’m on Path B.
I built six skills in 14 hours. I save $200/month. I own tools that do exactly what I need, with no usage limits, no pricing tiers, and no risk of some startup getting acqui-hired and shutting down my workflow.
And I’m shipping four SaaS products while working two day jobs, because I have the leverage to do it.
You can panic, or you can build.
I’m building.
If You Want to Start
Here’s the playbook I use every time.
Step 1: Pick Your Target
Start with something you use weekly but don’t need enterprise features for. The sweet spot is tools where you’re paying for convenience, not capability.
Good first targets:
Batch processing workflows (loop through files, process each, save outputs)
Video/audio transcription (Whisper is free and runs locally)
Content research and brainstorming (web scraping + Claude)
Grammar/style checking (Claude prompts replace Grammarly)
Format conversion pipelines (markdown → EPUB, video → transcript → summary)
Bad first targets:
Accounting software (compliance, integrations, audit trails)
Complex multi-step automation with 20+ triggers
Anything requiring hosted infrastructure you don’t want to manage
Real-time collaboration tools (Google Docs, Figma)
Step 2: Describe the Workflow, Not the Tool
Don’t say “build me a transcription tool.” Say “I have 20 course videos. I want to transcribe each one, distill the key points, and save them as markdown files organized by chapter.”
The more specific you are about your actual workflow, the better the tool fits. You’re not building a generic SaaS: you’re building exactly what you need.
Step 3: Start Ugly, Iterate Fast
Your first version will be rough. That’s fine.
My video transcription pipeline started as a janky script that choked on long files. I fixed the edge cases as I hit them. Now it handles 3-hour lectures without breaking a sweat.
Don’t try to build the polished SaaS version. Build the “works for my specific use case” version. That takes 90 minutes, not 90 days.
Step 4: The 80% Test
Run your homegrown tool alongside the paid one for 30 days. Track when you reach for the paid tool instead.
If your skill handles 80% of cases, cancel the subscription. The remaining 20%? Either iterate on your skill or accept the occasional manual workaround.
Perfect is the enemy of $X/month forever.
Step 5: Compound It
Once you’ve built one, you’ll notice patterns. The same techniques: file handling, API calls, text processing, output formatting: show up everywhere.
Your second skill takes half the time. Your fifth takes 20 minutes.
This is how you end up owning your entire stack without spending months building it.
The Prompt That Starts Everything
If you’re using Claude Code or similar:
I want to build a tool that [specific workflow].
I currently do this by [current manual process or paid tool].
My input is [what you're working with].
I want the output to be [format and destination].
What's the simplest way to build this?Then iterate. The AI will ask clarifying questions, suggest approaches, write code. You test, refine, test again.
Two hours later, you own something you would’ve rented forever.
The people who win the next decade won’t be the ones who worried about AI disruption.
They’ll be the ones who used AI to build tools, eliminate costs, and move faster than everyone else.
Don’t rent your stack. Own it.
P.S.: The AI wrapper economy is dying. Every “AI-powered” tool that’s just a UI around Claude or GPT is on borrowed time. The moment users realize they can build 80% of that themselves, the subscription revenue evaporates. If you’re building an AI SaaS, make sure your value is in the 20% that can’t be replicated in two hours.