317 Builders Just Shared Their Personal AI Tools. The Pattern Is Wild.

By Lazar Milicevic · Published June 9, 2026 · 7 min read

A user named aryamaan asked one question on Hacker News this week: what have you actually built for yourself since AI got good? The thread hit 317 upvotes and hundreds of comments in a day. I read every top reply, ran the patterns against what I ship for clients, and the signal is clear: the people winning right now are not buying SaaS. They're writing 100-line scripts that nobody else will ever see.

The Three Patterns That Show Up in Every Reply

If you scroll the thread looking for the next $200/month tool, you won't find it. What you'll find is a graveyard of tiny, ugly, hyper-personal scripts. Three patterns repeat in almost every top comment:

• Hyper-specific, not generic. Nobody built "an AI assistant." One person built a script that reads bank statement PDFs and categorizes expenses for taxes. Another built a Telegram bot that summarizes their parents' WhatsApp voice notes because of a time zone gap. Another wrote a CLI that watches ~/Downloads and renames PDFs based on content.
• Under 200 lines of code. Not because the builders are lazy. Because the moat moved. The hard part isn't writing the tool anymore — it's knowing which tool to write.
• 30 minutes to 4 hours saved per day. Per day, not per week. And the builders are mostly solo founders, freelancers, and small business owners, not staff engineers at Google.

That third number is the one that should make you uncomfortable if you're still paying $49/month for ten tools that almost solve your problem.

Pattern 1: "Boring" Personal Tools Beat Horizontal SaaS

The bank-statement categorizer is the cleanest example in the thread. Here's roughly what it looks like in production — I've shipped a near-identical version for a client running a small consultancy:

import pdfplumber, json
from anthropic import Anthropic

client = Anthropic()

def extract_transactions(pdf_path: str) -> list[dict]:
    with pdfplumber.open(pdf_path) as pdf:
        text = "\n".join(p.extract_text() for p in pdf.pages)

    msg = client.messages.create(
        model="claude-sonnet-4-5",
        max_tokens=4000,
        messages=[{
            "role": "user",
            "content": f"""Extract every transaction from this bank
statement. Return JSON array with fields: date, amount, merchant,
category (one of: software, travel, meals, payroll, other).

{text}"""
        }],
    )
    return json.loads(msg.content[0].text)

That's 18 lines. It replaces a bookkeeper doing 2 hours of manual categorization per month. Run it on a cron, dump the output to a CSV, hand the CSV to your accountant. Done.

The horizontal SaaS version of this costs $30/month, asks for full read access to your bank, makes you click through a dashboard, and still gets categories wrong because it doesn't know that "AWS" is your infrastructure and not "shopping." The 18-line version knows, because you told it.

Pattern 2: The Stack Is Almost Always the Same

Read enough of these scripts and you realize the entire AI-era personal-tooling stack fits on a napkin:

• A trigger (cron, webhook, file watcher, inbox poll)
• An LLM API call (Claude, GPT, or a local model)
• A sink (Telegram, email, Notion, a CSV, a Postgres row)

That's it. The Telegram-summarizer for voice notes is just:

# Triggered by Telegram webhook on new voice message
import requests, openai

def handle_voice(file_id: str, chat_id: int):
    audio = telegram_download(file_id)
    transcript = openai.audio.transcriptions.create(
        model="whisper-1", file=audio
    ).text

    summary = openai.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{
            "role": "user",
            "content": f"Summarize in 2 sentences, in English:\n{transcript}"
        }],
    ).choices[0].message.content

    telegram_send(chat_id, summary)

Whisper transcription runs about $0.006/minute. GPT-4o-mini summarization is a fraction of a cent. A 5-minute voice note costs ~$0.03 to summarize. The builder in the thread mentioned processing roughly 40 voice notes a month — that's $1.20/month to never sit through another rambling family update at 2am.

Compare that to the "AI family communication app" some YC startup will pitch you next year for $19/month.

Pattern 3: The Platform Tax Is Real and Compounding

Here's the math the thread implicitly makes. Ten SaaS subscriptions at $50/month is $6,000/year. That number is the floor. The actual cost includes:

• Context switching between dashboards (15-30 min/day)
• Integration glue that breaks every time a vendor changes their API
• Onboarding new team members on ten different UIs
• The features you pay for but never use (usually 70%+ of each tool)

Now compare to a custom stack. A small business I worked with this year replaced four tools — a CRM add-on, an email parser, a quote generator, and a scheduled-reports tool — with roughly 600 lines of Python running on a $6/month VPS. Total replacement cost: about 14 hours of build time. Ongoing cost: ~$8/month including API usage. Annual savings: $2,400 in subscriptions and roughly 6 hours/week of manual work that disappeared.

The platform tax doesn't show up on a single invoice. It shows up as "why is everything so slow" at the end of the quarter.

How to Find Your Own 100-Line Scripts

You don't need to read the HN thread to find your automation backlog. You need a notebook and one honest week of self-observation. Here's the exercise I give every client before we scope a project:

For 5 working days, log every task that:
  1. Took more than 15 minutes
  2. Felt like copy-paste between two systems
  3. You've done before and will do again

Format: [trigger] -> [you doing X] -> [result somewhere else]

Examples that come back almost every time:

• New lead email → manually copy fields into CRM → send templated reply
• Monthly invoice cycle → check hours spreadsheet → generate PDF → email client
• Weekly status report → pull metrics from 3 dashboards → write summary → Slack
• Inbox triage → read 80 emails → reply to 5, archive 70, flag 5 → repeat tomorrow

Every one of those is a sub-200-line script. None of them require you to be a senior engineer to scope. Once you can write the workflow as [trigger] -> [transform] -> [sink], you can automate the workflow — or hire someone who can do it in a day instead of a quarter.

The Hot Take Buried in the Thread

The era of horizontal SaaS for small businesses is ending. Not because the products are bad — some are excellent — but because the cost of building a vertical, owned, custom automation just collapsed by two orders of magnitude. The winners over the next three years won't be the companies with the biggest SaaS stack. They'll be the companies with five to ten small automations wired into their exact workflow.

The HN thread isn't a curiosity. It's the early signal. The 317 upvotes are 317 builders who already figured out that the moat is workflow knowledge, not code. The code is cheap now.

Why bizflowai.io helps with this

A lot of what I build for clients under the bizflowai.io banner is exactly this pattern: short, owned automations that replace specific copy-paste workflows — inbox triage routed to the right person, lead enrichment pulled from public sources into a CRM, invoice generation triggered by a timesheet row, weekly metric digests assembled from three dashboards. Each piece is small, each piece is yours, and none of them are a $49/month subscription that almost fits.

Frequently asked questions

What is the Ask HN thread about AI-built personal tools?

A user named aryamaan posted an Ask HN question asking what people have built for themselves since AI arrived. It hit 317 upvotes and hundreds of comments within a day. Responses revealed builders creating hyper-specific personal tools, like scripts that auto-categorize bank statements, Telegram bots that summarize WhatsApp voice messages, and CLIs that auto-rename PDFs based on content.

How much time do people save building their own AI tools?

According to the Hacker News thread, people building their own custom AI tools save between 30 minutes and 4 hours per day, not per week. Most of these builders are not FAANG engineers but solo founders, freelancers, and small business owners who built tiny scripts instead of waiting for a SaaS product to solve their specific problem.

Why does building custom automations matter for small businesses?

Paying for ten SaaS subscriptions at around $50 each costs roughly $6,000 a year, plus context-switching costs and broken integrations. This 'platform tax' compounds over time. Custom automations, often under 200 lines of code, can replace repetitive copy-paste work entirely. Businesses with five to ten custom automations wired into their workflow are positioned to outperform competitors relying on large horizontal SaaS stacks.

How do I identify tasks to automate in my business?

Open a note and write down every task you did this week that took more than 15 minutes and felt like copy-paste work. Examples include email triage, invoice generation, lead enrichment, and status reports. Anywhere you acted as human glue between two systems belongs on the list. That list becomes your personal automation backlog, with each item likely solvable by a roughly 100-line script.

When should I build a custom tool vs buy SaaS?

Build custom when the workflow is specific to you, repetitive, and currently requires manual gluing between systems. In the AI era, a Python script plus an API call plus a cron job can replace features that previously required a team six months to build. Buy SaaS only when the problem is genuinely horizontal and the tool fits your exact workflow without forcing expensive integrations or context switching.

Want more like this?

I publish practical AI automation, GenAI engineering, and faceless content workflows on YouTube every week.

Subscribe to bizflowai.io on YouTube — never miss a new tutorial.

Planning an AI automation project or need a second opinion on your architecture?

Connect with me on LinkedIn — Lazar Milicevic, GenAI Engineer & bizflowai.io Founder.

Visit bizflowai.io for our services, case studies, and AI consulting.