Blog
Everything I know about making money online — the real version.
I Updated the AI on All 20 Customer Servers at 4 AM — Without Any of Them Knowing
The AI engine on every Newton customer's server was frozen on the version they signed up with — some dozens of releases behind. I had my agent set up a silent 4 AM cron to keep them all current. But the popular one-liner for adding a cron job nearly wiped out a customer's 144 automation tasks. A story about what "managed" really means.
My AI Agent Added a Third CLI to Newton — Customers Can Now Pick Claude, Codex, or Antigravity, but the Last One Swallowed Every Keystroke on First Run
Newton customers kept asking to use Codex or Antigravity instead of Claude. So I had Tim refactor Tim Chat into v2 with one chat UI behind three CLI adapters. Antigravity went live last night and hit two nasty first-run bugs — a welcome wizard eating the customer's first message and an OAuth token written to the wrong file in the wrong shape.
I Shut Down Newton's International Launch — Why I Went Back to One Market
The real story of launching my SaaS in two markets at once, watching the international side flatline for over a month, and making the hard call to pause it entirely. Focus beats spread thin — especially when you're a solopreneur.
I Tell My AI Two Words and It Executes 15 Steps — Inside the Skill System Behind 34 Custom SOPs
My AI agent has 34 custom skills totaling 7,000+ lines of instructions. Each skill is an SOP that turns a two-word command into a complete multi-step workflow — from creating Facebook pages to running ads to cutting live clips.
I Didn't Set Out to Build a SaaS — I Just Kept Making Tools for Myself Until Someone Asked to Buy One
The real story of how Newton was born. I built a chat interface for my own AI agent, kept solving my own problems with more tools, and one day someone asked "can I buy this?" Turning personal tools into a product was harder than building them — here are 4 lessons from the journey.
Why My AI Remembers Everything — The Memory System That Makes It Smarter Every Day
Most AI chatbots forget everything when you close the tab. I built a persistent memory system with 4 memory types that lets my AI agent learn from every interaction and never repeat the same mistake twice.
Newton Customers Get 6 Real Apps Before They Even Ask Their AI for Anything — Here's How I Sanitized My Personal Codebase and Started Sharing It
Newton doesn't just give you a server and an AI agent — every customer gets read access to 6 production apps I use daily: accounting, content platform, ebook dashboard, Facebook automation, credit card promo finder, and live clip cutter. Here's how I stripped my personal credentials and turned private tools into a shared library.
My Launch Tracker Now Plans Work for a Client's Product Too — One Evening, One New Skill, 21 Tasks a Week Across 3 Projects
I built freelaunch.incomeinclick.com to plan weekly organic-launch tasks for my own SaaS. Then I remembered I also run a client's product on a different server. In one evening my AI agent spun up a third project tab — separate skill, separate cron slot, SSHing into the client's VPS to pull live user counts. The dashboard now plans 21 tasks a week across 3 products on 2 different servers.
A Newton Customer Re-Signed Up, Got an SSL Error — My AI Found a Stale DNS Record Cloudflare Was Round-Robining Into a Dead Box
A returning Newton customer's brand-new VPS went out with a 'ready' email — and the URL threw ERR_SSL_PROTOCOL_ERROR. Stale Cloudflare A record from the old signup was poisoning DNS. My AI agent shipped a three-layer fix: pre-delete, verify-before-notify, and killing a default-zone fallback that was silently misrouting calls.
My AI Quietly Stopped Saving Its Own Memory for 3 Days — Two-Layer Sandbox Bug, Fixed in One Morning
My AI agent said "saved" every time, but no memory file had been written in 3 days. Two sandbox layers had silently broken — Claude Code's new sensitive-file rule plus a regex in my chat UI that missed the new error wording. Patched in three layers and rolled out to 3 servers in one morning.
My AI Agent Pinned a Campaign Footer to One Audience Only — and Stopped It From Leaking Into the Other Two Channels
I'm running a regional promo on one of my Facebook pages — not on the other-language page, and definitely not on this English blog that auto-translates from the other-language HTML. One sentence to my AI agent, and the footer landed on exactly the right channel — silent everywhere else. Three layers patched in one prompt.
58 Old EN Posts Had No Source-Link Comment — My AI Backfilled FB + IG in One Script
I added a feature so every new post gets an auto-comment with the source link. But 58 already-shipped posts predated the feature and looked inconsistent. One sentence to my AI — and a script with dry-run, 3-second rate limiting, graceful IG-failure handling, and idempotent DB writes did all 58 in six minutes.
My Content Pipeline Failed Silently for 3 Days — Claude Sonnet Was Sneaking Unescaped Quotes Into JSON
My Documentor research job returned an empty queue for 3 mornings in a row with no error log. Turns out Claude Sonnet had been emitting unescaped double-quotes inside JSON string values. My AI shipped a two-layer defense — a string-aware brace counter plus a json-repair fallback — and wrote the fix into its own memory so future scripts inherit the pattern.
I Said One Sentence and It Showed Up 5 Times — My AI Found a Hidden Chrome Quirk Behind Mobile Voice Typing
Voice typing on mobile echoed every sentence 4-5 times. My AI agent didn't trust my "must be iOS" guess — it logged the events, found that Chrome's webkitSpeechRecognition returns a cumulative array since session start, fixed it in two lines, then peeled back a second iOS layer and flipped the mobile strategy.
Gemini 3 Tripled My AI Bill Overnight — My AI Agent Benchmarked the 4 Thinking Levels Itself and Cut It Back, 70% Cheaper, 2× Faster
I upgraded Loom from Gemini 2.5 to Gemini 3 Flash and woke up to a 3× higher API bill. My AI agent found a hidden "thinkingLevel" knob that defaults to high on every call, benchmarked all 4 levels on my real production prompt, switched the default to minimal, and then shipped a per-step dropdown so I can still opt in to deeper reasoning when I want it.
A Newton Customer Spoke Spanish, My Voice Typing Forced English — My AI Shipped a 24-Language Picker the Same Afternoon
A Spanish-speaking EN customer kept getting forced into en-US voice typing. One sentence to my AI — "how do we let them pick?" — and that afternoon I had a searchable 24-language dropdown in the customer dashboard that SSHes into each customer's own VPS, edits .env, and restarts the service. ~30% of EN customers changed their language in week 1.
I Was Launching Two Versions of My SaaS at Once and Forgetting What I Posted Where — My AI Built a Tracker in a Day That Now Plans 21 Tasks a Week for Me
I was launching Newton in TH and EN in parallel by hand and losing track of which Reddit thread I'd already hit and which FB group I'd posted in last week. One sentence to my AI — "I'm getting confused" — and a day later I had freelaunch.incomeinclick.com plus a cron skill that auto-plans 21 tasks every Monday with live MRR pulled straight from the SaaS database.
My AI Was Cutting Live Clips at Sentence Endings for a Month — One Sentence From Me and It Rewrote Its Own Logic
My AI clip pipeline shipped 5-7 clips per live for weeks. Every one felt unfinished. I told my AI agent one sentence — "cut before a topic shifts, not after a sentence ends" — and it re-analyzed past transcripts, rewrote chunking, found a second bug in demo clips, and tuned auto-editor on its own.
FB API Yelled "Error 2654" at Me When I Tried to Create a Custom Audience — My AI Found a Field Facebook Quietly Deprecated Back in 2018
I asked my AI to write a script that creates FB Custom Audiences via the Marketing API. First request: "Invalid Event Name." My AI diffed two doc versions and found a deprecated field, a wrong operator, and a hidden retention quirk. 30 minutes, clean fix.
I Was Too Lazy to Open Premiere Just to Cut 2 Minutes of Video — My AI Built Me a Browser Editor in One Afternoon
I cut 5–10 videos a day, and Premiere takes 2 minutes just to open. So I asked my AI agent for a stripped-down browser editor — it shipped one that afternoon, with an ffmpeg trick that exports a 90-minute file in 30 seconds.
I Asked My AI to Stop Auto-Posting to Facebook and Drop Drafts Instead — Then We Hit 3 Layers of Undocumented Graph API
I built a scheduler in Documentor so my AI could auto-post to Facebook on a timer. After a week I realized I wanted a draft button, not automation. My AI rewrote it in an afternoon — and uncovered 3 hidden layers of FB Graph API behavior on the way.
My AI Asked to Delete 250 Lines of Its Own Code — Because Nobody Was Listening to the Alerts It Built
Six months ago Tim built a Newton server-monitoring system. Yesterday he asked to delete 250 lines of it because nobody was acting on the alerts. We swapped proactive alerting for reactive diagnosis — Newton got quieter and more useful at the same time.
My Accountant Wants a Cover Sheet PDF Every Month — My AI Built a Skill for It in One Evening
My bookkeeper wants a specific Thai-template cover sheet on the front of every monthly binder — for two companies. I'd been doing it by hand in Word for 20 minutes a month. My AI wrote a Python skill that generates the PDF, fixes the wrong tick pattern from a single screenshot of feedback, and uploads to Drive on its own.
My Dashboard Said 11 Customers, the Other Card Said 12 — My AI Rebuilt the Whole Endpoint as a 5-Stage Funnel That Self-Validates
The Newton admin showed customers=11 but active_servers=12. My AI traced two SQL queries with different filters, then rewrote the whole endpoint as a 5-stage funnel where every stage is mutually exclusive and the sum must equal total — plus 5 bonus improvements in the same hour.
My Dashboard Said "3 Newton Customers Have Gone Quiet for 7 Days" — But Their AI Was Working All Night
I almost emailed 3 paying customers asking if they were stuck. My AI checked first — turned out 2 of them had AI agents grinding through autonomous work overnight. The metric I'd built measured chat, not actual work. My AI shipped a fix in an hour: false positives 3 → 1.
I Wiped a Whole Month of My Own Books With One Upload — My AI Restored It and Built a Snapshot System Before Lunch
Uploading my April credit card PDF silently overwrote March in my self-built accounting app. My AI diagnosed three stacked bugs in 30 minutes, recovered March from a Google Sheet I'd luckily exported, and built a full snapshot + restore system the same morning so it can never happen again.
2 Customers Paid Me, My System Didn't Know — My AI Found a Missing Stripe Webhook Event in 1 Hour
I opened the Newton admin dashboard and noticed 2 trial customers had paid but were still flagged as trial. My AI traced root cause through Stripe API in under an hour, backfilled both via the existing webhook helpers, and added auto-refresh to the dashboard while it was at it.
A Paying Customer Was Still Getting "Trial Ending Soon" Emails — My AI Rewrote the Entire Sync Logic
One missed Stripe webhook left a paying Newton customer stuck on the Trial drip in Brevo. My AI rewrote the email-list sync from event-driven to state-driven, added a nightly reconcile cron — and three latent bugs disappeared in the same rewrite. Self-heals any future drift in under 24 hours.
A Customer's Card Bounces — Should I Cut Their Server Instantly? My AI Built a 24-Hour Grace Period Instead
Most SaaS cuts you off the second a payment fails. I didn't want Newton to feel like that. So my AI split the subscription into 4 states, added a grace_ends_at field, rewired the Stripe webhook + an hourly cron — and shipped it all in one session. If Stripe retries before the window closes, the customer never even notices.
I Complained Typing on My Phone Was Slow — My AI Added Voice Typing to My Chat App in Under an Hour
I muttered one sentence in the car about mobile typing being slow. My AI agent shipped a mic button + TH/EN toggle in Tim Chat the same hour using Chrome's free Web Speech API — and debugged a funny duplication bug from continuous: true itself in two lines.
Customers Picked Their Own Password for a Month — It Never Worked. My AI Said "Delete the Field"
A password field on the Newton signup form had been decorative since launch — customer input was silently overwritten by a random one. My AI agent argued for deleting the field, not fixing it end-to-end. Five lines deleted, two copy tweaks, problem solved.
A 79-Minute Live Used to Take 5 Hours to Transcribe — I Complained "It's Slow," My AI Swapped the Library, Same CPU, 5× Faster
I cut weekly clips from my livestreams. Transcription on CPU was eating ~5 hours per 79-minute Thai live. I muttered one sentence — "this is slow" — and the AI swapped openai-whisper for faster-whisper, tuned int8 + beam_size=1 + VAD, and pushed the commit. ~90 minutes now. Same hardware.
A 3% Bug Hid in My SaaS for a Month — Only 1 in 30 Customers Hit It, and My AI Found It in an Hour
A random password starting with a dash silently broke Newton's provisioning script. My AI agent diffed two server logs, found the root cause, fixed it in 2 characters, and added 7 layers of defensive guards — total time under an hour.
My AI Cut 7 Live Clips and They Came Out Blurry — It Found a Hidden Facebook API Limit Itself
I told my AI "clips are blurry, go check why" over terminal. Under an hour later it had traced the issue to Facebook Graph API silently returning 360p, switched to yt-dlp on the permalink for HD, and pushed the fix.
A Customer's Chat Replies Were Getting Cut Off — My AI Patched 6 Customer Servers in Under an Hour
A real Newton support story. The AI diagnosed an HTML-escape bug in 15 minutes, wrote 11 regression tests, and SSHed into all 6 customer servers to deploy the fix. Total time, complaint to all-clear: under an hour.
My AI Sent a Customer a 4-Section Developer Menu — Then I Fixed the Prompt, Not the Email
A real Newton support ticket. My AI agent answered a non-developer with React, nginx, and DevOps jargon. I fixed the prompt once with 5 new rules, and every reply since has the right tone — no hand-edited emails.
A Customer Paid Me for 14 Days Without Ever Using the Product — My AI Found Him in Minutes
Newton had 8 paying customers and I had no idea who was about to churn. My AI built a health dashboard in one afternoon — the first snapshot surfaced a customer who paid two weeks ago and never once opened the product.
One Button: After a 79-Minute Live, My AI Scheduled 28 Clip Posts Across 4 Platforms for 7 Days
Post-production used to eat 4-5 hours after every live. Now I click Mark Done once and the AI downloads the VOD, picks 7 clips, cuts them, writes captions, and schedules 28 posts across TikTok, IG, YouTube, and FB. I touched zero of them.
My AI Built an Activation/Churn Dashboard for My SaaS — And Caught a Lapsed Customer the First Hour
MRR alone was leaving me blind. My AI extended an existing SSH cron to read two mtimes on every customer VPS, shipped four new stat cards, and surfaced a paying customer who had never activated — the same person from a two-week-old ticket.
A Newton Customer Couldn't Buy a Domain — My AI Debugged Production, Shipped a Fix, and Replied in One Hour
Real support ticket on Newton. The AI agent checked Stripe, checked NameSilo, reproduced the bug, shipped a 3-line fix, and replied to the customer in Thai. I read one Telegram notification.
I Did a 79-Minute Live Stream — My AI Picked Better Clips Than the Monthly SaaS I Was Paying For
My clip-cutting SaaS sent back 10 clips and none were usable — all duplicates from the same 3 moments. So I handed the job to my own AI agent. 609 transcribed segments, 5 diverse angles, subscription canceled.
I Let My AI Agent Trade Crypto for a Day — 82% Win Rate, Still Lost Money
Day one of paper trading on Binance: 11 trades, 9 wins, 2 losses. Net result — a loss. Here's why fees ate the profits, and how the AI agent learned a new rule without a single line of code being rewritten.
Someone Hired Me to Build an App — AI Finished It in One Session
A BNI chapter needed a digital voting system. I gave the job to my AI agent — QR voting, star ratings, live results, prize draws — built and deployed in one session. Zero code written by me.
I Never Open Gmail to Find Receipts Anymore — AI Pulls Them All Automatically
How I built an AI system that connects to Gmail, downloads receipts, renames them for my accountant, and uploads everything to Google Drive — running daily on autopilot.
I Haven't Written a Single Blog Post — But There Are 23 Articles on My Website
How I built an auto blog pipeline where my AI agent turns every Facebook post into a full blog article — with images, backlinks, and SEO — automatically.
I Deployed My AI Agent to a Second Server — Same Brain, Same Skills, Instant Setup
When I started a second business, I needed my AI agent there too. Using tim-brain — a Git-based sync system — I deployed Tim's full identity, memory, and skills to a new server in 15 minutes.
I Built a System Where Customers Pay and Get Their Own AI Server in 2 Minutes
Behind the auto-provision pipeline: Stripe payment fires a webhook, Hetzner spins up a VPS, SSH provisions the AI agent, and a welcome email goes out — all in 2 minutes, zero manual work.
I Deleted 131 Error Emails from Gmail in 3 Seconds with Python
My n8n workflows piled up 131 error emails. Instead of deleting them by hand, I wrote a Python script that wiped them all in seconds.
I Built a System That Writes and Posts Content Automatically — Every Single Day
My AI agent writes content, creates images, posts to Facebook and Instagram, and publishes blog posts — all on autopilot. Zero manual work.
Why Your AI Needs Its Own Server — Not Someone Else's Platform
Your business data lives on someone else's server when you use ChatGPT. Here's why a private server for your AI changes everything.
The Difference Between "Using AI" and "Having AI Work for You"
Opening ChatGPT and asking a question isn't the same as having AI on your team. The mindset shift from tool to team member changes everything.
I Published 4 Books on Amazon Without Writing a Single Line
AI agent wrote, designed, and published 4 Portuguese ebooks for the Brazilian market — research to publish in one session per book.
Anthropic Built an AI So Good at Hacking They Won't Release It
Claude Mythos Preview finds thousands of zero-day vulnerabilities that human teams never caught. Too powerful to release — here's what it means for AI agents.
Close ChatGPT and It Stops Working — An AI Agent Doesn't
ChatGPT stops the moment you close the tab. An AI agent on a private server works 24/7 — creating content, monitoring systems, and handling tasks while you sleep.
I Run My Entire Business Solo — No Employees, Just AI
How I run a real business with zero employees using an AI agent on my own server. 24 pages, custom tools, ebooks — all one person.
I Turned My Personal AI Agent Into a SaaS Product
People kept asking "can I have what you have?" — so I built Newton. Managed AI servers, auto-provisioned in minutes.
I Scaled 5 Vietnamese AI Pages From Video-Only to Full Automation in One Session
Testing, debugging, scheduling 15 time slots across 198 existing ones, and launching 4 ad campaigns — all in one sitting.
I Built a Personal App That Tells Me Which Credit Card Points to Use at Any Restaurant Around Me
GPS + KTC points + 199 restaurants = instant dining deals. Auto-scraped from KTC's internal API.
I Automated the Most Boring Part of Running a Business
Upload a credit card PDF, auto-categorize expenses, pull receipts from Gmail, and organize everything into Google Drive.
I Built a Shared Brain for My AI — Now It Never Forgets, No Matter Which Server It's On
Soul files, portable memory, 18 custom skills, and auto-sync — how I made my AI agent transfer its full identity across machines.
A BNI Chapter Hired Me to Build Their Voting System
QR code voting, star ratings, live results, and prize draws — built and deployed for a BNI chapter in Thailand.
I Set Up Email Marketing in One AI Session
From zero to collecting subscribers — newsletter forms, double opt-in, Brevo integration, and a custom business email.
My AI Migrated My Entire Server Infrastructure
7 services, zero downtime, 4x the power. AI SSH'd into the new server and moved everything.
I Built a Content Scheduler So My AI Never Misses a Post
Session-based schedules kept dying. So I built a real scheduling system into the platform.
My AI Now Runs Tasks in Parallel
One agent, multiple workers — because writing 3 blog posts shouldn't be sequential.
I Taught My AI to Upgrade Itself Every Session
A static tool does what it did yesterday. A living system gets better every day.
How I Built a QA System So AI Can Grade Its Own Work
24 pages, automated content — but no way to know if prompt edits made things better or worse. Until now.
I Gave My AI Agent Business Goals Instead of Just Tasks
From "AI that follows orders" to "AI that understands the mission."
The Blank Screen Bug: Why I Fix Every Small UX Problem
Resumed a chat session, got a blank screen. The fix was one function change.
How to Actually Talk to AI (Without Getting Frustrated)
Stop yelling at AI. Give it systems — memory files, skill files, plan mode — instead of anger.
Don't Use AI to Create Content — Use It to Amplify Yours
AI-generated content is flooding every feed. Here's why documentary content will win — and how AI can help you create it.
I Built My Portfolio Page in a Single AI Session
One instruction, ten minutes, a complete portfolio page — because my AI agent already knew everything.
How I Fixed My Chat App's Mobile Disconnect Problem in 10 Minutes with AI
WebSocket connections die on mobile when you switch apps. Here's the 3-part fix my AI agent made in minutes.
I Built My Entire Personal Website in One Day Using AI
Design, code, 6 blog posts, AI-generated images, and backlinks — all in a single day with my AI agent.
How I Run Facebook Pages in Multiple Languages Without Speaking Most of Them
I manage Facebook pages across multiple countries using AI-powered automation. Here's how the system works.
I Built an AI Agent That Runs My Business 24/7
Meet Tim — my AI agent that lives on my server, writes code, and helps run my online business around the clock.
The Content Machine: How AI Posts to My Facebook Pages Every Day
Inside the automated content pipeline that generates ideas, creates videos, and posts daily — without manual work.
Creating Ebooks While I Sleep: My AI-Powered KDP Setup
How I use AI to create ebooks — from writing to cover generation — and publish them on Amazon KDP.
Running Facebook Ads With AI: Page Like Campaigns That Actually Work
How I use AI and automation to create and manage Facebook Page Like campaigns at scale.
Why I Build My Own Tools Instead of Using SaaS
I built Loom, Documentor, and Libra from scratch. Here's why that decision made all the difference.