Hey builder,
For the last three years the AI conversation has been dominated by one archetype: the well-funded founder, the ex-FAANG engineer, the full-time creator who quit their job to go all in on AI. That story sells a clean version of what it takes to build right now. A runway. A team. A quiet room.
The real story is messier. Most of us are building between shifts. We are hitting weekly rate limits before Wednesday. We are uploading developer-account ID photos from the front seat of the car between jobs. We are choosing between upgrading the Claude plan and getting the back window of the car fixed, because someone smashed it on a Tuesday morning and now both bills are due.
Today I am breaking down what it actually looks like to ship AI tools from inside a 9-to-5 and a full family life. What is in the stack. What it costs. What breaks. And the three tools I have built that are already buying back the very hours I spent building them. Call it the Budget Builder's Playbook.
Today's Author
Prof. J is a builder, taxi driver, host of the AI Builders Showcase, and co-host of weekly AI Spaces on X. Currently shipping Prep2Eat, LocalGlowUp, ACC Network, and a stack of agents and bots, all on nights, weekends, and the spare minutes between fares.
The Budget Builder's Playbook
How I'm shipping AI tools from a taxi, on a 9-to-5, with a family.
Everyone selling AI as the great equalizer is half right. Yes, a laptop and a Claude plan now puts capability in your hands that a $50M Series A could not buy two years ago. That part is real, and it is the reason I keep building.
The half no one writes about is the other side of the equation: the rate limits, the API spend that creeps up every time you add a project, the eight-year-old laptop choking on a 1.6 MB Android build, and the simple math of trying to ship anything between a 6am taxi shift, a 4pm family dinner, and the 10pm bedtime that is supposed to be non-negotiable but never quite is.
What I have learned over the last eight weeks of journaling every build, every hit, every miss: the constraint is the curriculum. Building on a budget, on stolen time, with second-hand hardware, forces you to build the right things. The things that pay back time, not the things that look impressive on X. This is what the math actually looks like, and the three tools I built that are already turning that math in my favor.
Section 1. The Stack
What I'm actually paying for
Most builders couldn't tell you what their AI stack costs in a given month because they have never added it up. That is not an accident. It is death by a thousand $20/mo charges, and the first rule of the Playbook is add it up. Here is mine, current as of this week.
AI models and coding tools
- Claude Max plan. Upgraded from Pro after the Pro tier started capping me mid-week, and Pro itself was an upgrade off the free tier in late March when Cursor's limits got brutal.
- Codex / ChatGPT plan. Separate subscription. Hit weekly limits the same day I launched the first newsletter issue.
- Cursor. Still paying for it, mostly because some older projects (BaseWalk, Poly Agent, Copy Trader Bot) were built in it.
API keys (each one a small monthly burn)
- 7 separate OpenAI API keys, one per project (Prep2Eat, Base Agent, Spaces Dashboard, WEB3TLDR, Podcast Summarizer, Sora AI Influencer, and the main key)
- 2 OpenRouter keys for OpenClaw and Hermes agents
- 2 HuggingFace keys for HF Space deployments
- TradingView token (currently expired, a chore I keep deferring)
- Twitter/X API set, Calendly, Browser Use, Birdeye, Gaia, ODDS API, API-Football. Every one of them small. Every one of them adds up.
Hosting and infra
- Hostinger VPS. Runs the AI News Bot and the Copy Trader Bot in Docker, also hosts the nightly cron that calls me every night (more on that below).
- Vercel. LocalGlowUp and ACC Network.
- Laravel Cloud. Prep2Eat.
- Supabase (×2 projects, free tier), MongoDB Atlas (free tier), Upstash KV (free tier).
- Spaceship for domains.
- MailerLite for newsletter sending (this very article was drafted there first).
- Vapi for voice on the Accountability Agent.
One-time fees that hit hard when they hit
- Google Play Developer account: $25 (paid 4/23. Uploaded my ID from the car between rides.)
- Apple Developer account: $99/yr. Haven't paid this. Can't yet. That is why Prep2Eat is going to Android first.
- Android Studio install required me to clear disk space on a laptop with no headroom before I could even start the Prep2Eat submission process.
The honest total
$250 to $300 a month
In pure subscriptions, before any API usage spike. Yes, I upgraded the Claude plan. The math got worse before it got better, and that is the actual answer to the question nobody else writes down.
Takeaway you can use today
Before you add one more tool, open a spreadsheet and list every AI-related charge hitting your card. Sort by "do I use this every week." Kill anything below the line.
Section 2. The Drag
What actually slows the build down
Here is what nobody puts in the build-in-public posts. Some real entries from my journal over the last four weeks.
- 4/13. "Day started off slow as I went to my car for work, someone broke my back window and had to get it fixed. Time and money wasted."
- 4/20. "Ran into Claude limits for the week. Can't use until 4/23 unless I upgrade. Looks like I'll be joining the Max plan real soon."
- 4/21. "Worked overtime to fund Claude plan upgrade and developer accounts (Google Play + Apple)."
- 4/27. "Horrible! Got a little work done but got blocked with hardware constraints. Spent a lot of time on the road for work."
- 5/2 to 5/3. "Just work, no code or AI. Gotta catch up on some bills and other expenses."
- 5/9 to 5/10. "Just work and time with family. Mother's Day and big karate tournament for my daughter."
- 5/12. "Launched first issue of newsletter, got about 6 subscribers but then hit limits for the week on Codex."
- 5/13. "No work done today. Had to attend a funeral and spend time with family."
That is a one-month window. Inside of it: a smashed window, a hardware-blocked weekend, a funeral, a Mother's Day, a karate tournament, two separate rate-limit walls, and a catch up on bills pause.
Anyone telling you they are building AI side projects without weeks like this is either funded or lying.
The reframe that finally clicked: the drag isn't a bug in the plan. The drag is the plan. The whole point of building leverage tools is that they keep running on the days you can't.
Section 3. The Payoff
Three tools already buying back my week
This is where the math starts turning. Three builds, three different jobs, all running right now.
Trading Army
Built across 4/15 to 4/17 in three of the hardest sessions I've ever logged. The 4/15 journal entry literally says "THIS WAS HELL." Six separate Windows issues (WindowsApps permissions, MSIX package path changes, IPv6 routing, cloud-vs-local MCP architecture) before I could even get a TradingView chart to read into Claude.
What it does now: I type claude from the TradingArmy folder and a live trading analyst spins up. It reads my chart, runs Brinks Box analysis, identifies RVC and GVC formations, and enforces all 97 strategy rules I extracted from the eight Traders Reality checklists. When I say journal this, the trade gets written directly to my Excel journal. Auto-formulas, dropdowns, the whole thing.
The payoff
What used to be 30 to 40 minutes of chart prep and post-trade journaling per session is now a conversation that ends when I close the window. The goal I set for it (50 to 100 trades through the system) is going to land months earlier than it would have without it.
AI News Bot
Python + Discord.py + Docker, running on the same Hostinger VPS that runs my copy trader. Fetches AI RSS feeds, dedupes against a SQLite database, posts a categorized daily summary to a Discord channel. I can hit !force_news and get an immediate refresh whenever I want.
The meta-angle: this is the bot that fuels the newsletter you may have already read. I updated it from my phone on 4/11 while on the road working, then finished the deploy in VS Code at home that night. It has been running 24/7 since.
The payoff
I no longer do a news pass before writing. The news pass is happening in the background while I drive, while I sleep, while I am at a karate tournament. When I sit down to write the issue, the source material is already curated.
Riley, the Accountability Agent
Riley is a Vapi voice agent built on 4/15. She calls me every night at 9pm EDT with a check-in script I wrote: what did I commit to today, what did I actually ship, what is tomorrow's one thing. The call ends, the transcript gets dropped into Google Drive via MCP, and Claude reviews the entry against my weekly goals doc.
The build itself has a moment I am proud of. I originally wired the scheduling through Zapier and hit a paywall almost immediately. On 4/18 I rebuilt the trigger as a Hostinger VPS cron job, DST-aware, fires nightly without my PC being on. Killed a recurring subscription and made the system more reliable in the same week.
The payoff
On the days the drag wins (the funeral, the karate tournament, the bill-paying weekend) Riley still calls. The system doesn't lapse. Note for the next iteration: I need to make her voice tougher. Right now she is too polite when I miss.
Section 4. The Rules
Five things I'd tell anyone starting where I started
Build leverage before you build features. The first three things I'd build in any new stack: the tool that logs what I did, the tool that calls me to keep me honest, and the tool that gathers tomorrow's inputs while I sleep. Trading Army, Riley, and AI News Bot were all leverage builds. Everything else gets easier once those exist.
Add up the stack monthly. Out loud. To yourself. $20 here, $25 there, an API key you forgot you minted. Kill anything you didn't actually use this week. The discipline matters more than the dollars saved.
Use the rate limit as a deadline, not an obstacle. When Claude or Codex cuts me off mid-week, that's the signal to switch to the other one, or to step away and ship what I already have. Limits force prioritization. Treat them like meetings.
Ship to one store, not both. Prep2Eat is going to Google Play first because Apple costs $99/year and Google costs $25 once. Pick the cheaper distribution path and learn there. App Store can come after revenue.
Your family is not the obstacle. Your hours are. The funeral, the karate tournament, the Mother's Day dinner. Those are non-negotiable, and they should be. The thing you can actually budget is the 6pm to 10pm window. Protect that window like it is billable, because in the long run it is.
Your Move
If this resonated, and especially if you are building something similar from inside a job, a family, or both, here are the three things I would ask:
- Forward this article to one other builder who is grinding on the same problem. The whole point of writing it was to put words around what most of us are doing alone.
- Follow along on X at @mommasboi89. I post the wins and the losses in real time.
- Subscribe to the newsletter if you want the Sunday Deep Dive in your inbox each week. Sign up here.
What else I'm building
Prep2Eat. Meal planning web app. Android submission imminent, waiting on Play Console ID verification.
LocalGlowUp. Local-business web agency. AEO-optimized site live at localglowup.com.
Spaces Dashboard. Crypto and AI news dashboard that powers my weekly X Spaces prep.
Marketing Agent OS. Multi-agent system being built to handle campaigns and growth across all of the above.
Built between fares. See you next Sunday.
git commit -m "shipped"