To figure out how to navigate this new era of agentic vibe coding, where everyday degens prompt AI to build APPs using natural language rather than writing syntax line by line (“Yo Claude, make this shit better”), I set myself a challenge:
Build something useful
I had three rules:
It must solve a real problem.
It must leverage Lindy-proven infrastructure (I absolutely do not want to vibe-code smart contracts).
It must be something I am passionate about.
The result is @tipz_cash
In four weeks, I planned, designed, built, iterated, audited, and launched (with a full GTM-content strategy) a fully functioning private tipping payment protocol with no coding experience.
But before you make your hasty judgments (oh shit, here we go again, more ai slop), I implore you to read a little further.
Having survived 5 years building in DeFi, I am not naive to the risks of security vulnerabilities, and having been on X for the past few months, I am definitely not naive to the saturation of shitty AI slop. So yes, I took rigorous measures to ensure a balance of both relentless efficiency and uncompromising safety.
And no, this article was not written with AI. And yes, I relied on some help from some gigabrained dev friends.
Let’s begin.
What is TIPZ?
tipz.cash is a private tipping protocol for creators and cypherpunks.
Zcash is the most battle-tested private payment technology. Zero-knowledge proofs, shielded transactions, and cryptographic guarantees that have held up under fire for nearly a decade. But it has always lacked a social layer - real, frictionless applications that bring private money to everyday use cases.
TIPZ fills that gap.
You connect your X account and generate a unique tipz.cash/@username payment link in less than a minute. Fans click your link, select an amount, and tip you in any cryptocurrency from ETH, USDC or SOL, which is then automatically routed and swapped into shielded ZEC.
Because of Zcash’s shielded addresses, your wallet balance stays private.
The Problem It Solves
TIPZ leverages one undeniable PMF of digital assets: borderless, peer-to-peer micro-transactions.
But it solves the two massive bottlenecks holding the creator industry back:
Creator Serfdom. Web2 platforms bleed creators dry through huge platform fees, supporter profiling, arbitrary deplatforming, and greedy middlemen.
Public Ledgers. If someone sends you a tip via crypto (which avoids the above), they can immediately see your entire wallet history and total net worth.
TIPZ fixes both by enforcing:
Self-custody.
Peer-to-peer routing.
No middlemen.
No deplatforming.
0% platform fees.
The hardcore cryptography and infrastructure already existed. What was missing was the UX layer, purpose-built for creators (an industry growing exponentially).
AI augmentation allowed me to bridge that gap and build the solution in weeks instead of months.
Compressing 6 Months into 4 weeks
Cross-chain payment rails, conversion optimised UI/UX, acquisition flywheels, funding strategy, and the entire GTM strategy were compressed from a half-year timeline into a few weeks using agentic AI.
To streamline distribution, I also spun up a bespoke Content Hub Engine (basically a CMS X Typefully on steroids). The thread/article/tweet creator, calendar, and visual asset social media bank (that can generate new brand-aligned visuals with a single prompt) were architected entirely between me and my terminal in one hour.
It does not automate marketing with generic AI verbose (it does not write content at all); it instead acts as a visual curation engine. It scours current market trends, researches the underlying tech advancements, and extracts highly relevant quotes, generates custom visual assets based on a specific tag (education, quotes, bangers), and pipes them directly into the content library visual bank for me to then curate into tweets/thread/articles.
Although just a side project to solve a single problem within the greater system, this streamlined one of the biggest bottlenecks I have experienced when executing high-velocity marketing projects; aligning market narratives and visual identity without sacrificing the soul of the brand.
Lessons from vibe coding a fully functioning payment app
So, what were the key lessons learned in the past 4 weeks?
1. Plan, plan, plan. Plan your architecture before you build it. Don’t just rely on the AI to do your research for you. If you have a deep understanding of a subject, you know exactly where the high-signal conversations are happening. Gather that relevant context yourself, ask questions, iterate, and manually feed that research into the AI before you start generating code. While a build gives that instant dopamine spike, rewire your brain to crave slow, thoughtful ideation. Know what you want, and articulate it explicitly.
2. OpSec is key. Granting an autonomous AI tool like OpenClaw full terminal access is an Operational Security nightmare. Realise that leaving backdoors open in your environment means anyone can walk in, and they aren’t always the people you want in your home. I recently locked everything down with trailofbits claude-code-config.
3. Structuring the AI’s authentic brain (via custom skills). Instead of throwing generic prompts at a blank window, I anchored the AI’s context using Claude Code’s Skills architecture. I created specific skill files loaded with core mental models from my favorite books on physics, system design, strategy, business, design, and creativity, as well as insights from experts in various fields. While attacking specific problems, I essentially moulded the AI’s brain to be the exact people I would want to ask for help with. The AI will source an array of information at the speed of technology, but it will be generic and powered by the AI’s own algorithm. To encode your authenticity into the system, you must strategically build your own databank of information. Just like your brain has been moulded by those you think highly of, mould your AI to source skills from an all-star “gigabrain” team of wisdom-seeking experts. Except this time, you don’t need to spend a week absorbing 10% of the information from a book; your AI absorbs, summarises, and implements 100% of the information in 1 minute.
4. Learn on the go (the AI is the teacher). The absolute best way to learn about AI is to just go and use it. The best teacher of AI is the motherfucking AI itself. Ask it questions. Be curious. If it tells you it can’t do something, don’t settle for it, keep prodding. You will begin to notice your brain changing the way it approaches a problem. The game is no longer “Solve for X.” It becomes, “What are all the different ways I could solve for X?” Play around, see what happens, and you will surprise yourself.
5. Bottlenecks are build opportunities. Every time you hit a bottleneck, that is an opportunity to solve a problem. When I hit friction in my marketing workflow, I asked Claude for three different approaches to optimize it, forcing it to give me the benefits and drawbacks of each. That exact process is how I ended up building the bespoke content hub. It is already way better than current products because I can seamlessly add new custom features whenever I want.
6. Avoid scope creep. The AI can build you anything you want, and it is dangerously easy to get carried away by it. Focus on a single task and solve that first. With the aforementioned content hub, I could have built a fully functioning SaaS application - a public, superpowered version of Typefully. But that wasn’t what I needed. I needed something custom that integrated my current local files and worked seamlessly from my terminal. I didn’t need API keys or database storage; I just needed a custom system that worked for me. Don’t get carried away. “You can do anything... but not everything.”
7. You don’t need a god-tier setup. Don’t get sucked into thinking you need the craziest computational rig with max API access and models running 24/7. I didn’t use OpenClaw (I’ve never used it), and I didn’t buy a Mac Mini (shock horror!). Start slow, iterate. If you run out of tokens, then think about increasing. I am running on the Claude Max $200/month spec, and have never run out of tokens. And in that same time, I haven’t just built TIPZ, I built an interactive portfolio website (in bio), a custom Bloomberg terminal, other business websites, and have been working on a full-stack APP for mountain navigation (weather, avalanches, climbing, flying), and a range of other apps (avoid scope creep he says...)
8. Iteration and the “centaur phase”. Contrary to popular belief, actual engineers are going to be some of the most sought-after people in the next year. Everyone needs a managerial overseer for their code projects. We are currently in the “Centaur Phase”, just like in chess, where a human-AI hybrid team could beat both the best solo human and the best solo supercomputer. Sure, we will eventually reach a point where the engine is always better on its own, but for now, if you don’t have a deep understanding of a system, use someone who does. I reached out to a trusted backend and frontend engineer to gut the slop, and then distilled their specific knowledge directly back into the AI to compound its intelligence. This wasn’t optional. On first review, Franz found API endpoints with zero authentication and sensitive data stored in plaintext, completely exposed. The AI had built something that worked, but was wide open. One security review later, everything was locked down. I fed his fixes and notes back into the Claude MD file, and the next iteration was dramatically better. If you’re building anything that touches payments or user data, get a human to review it. The AI doesn’t know what it doesn’t know.
What comes next?
We are living in a weird timeline. And honestly, a lot of it is rather scary. AI overlords, widening deficits, political unrest, civil uprising, mass surveillance, and geopolitical financial control. We are granting more power to the powerful, but with it, we are distilling an unstoppable power in us all: the power to create, the potential to impact, and the ability to drive real change.
A month ago, I couldn’t write a frontend to save my life. Yesterday, I shipped a fully functional, cross-chain privacy protocol, albeit leveraging proven, heavy-hitting infra like @Zcash, @zodl_app, and @near_intents.
But my point is this: Every day, the boundary of possibilities is extending.
It is like LEGO evolved into the most badass form possible. Instead of plastic bricks, it is software, applications, and websites. All you need to do is identify the foundations you want to leverage and distill what you want to build into a chat box, and voilà, it appears (after a few iterations).
So, what comes next?
I don’t know. But I implore you to build.
Solve a real problem.
Leverage proven infrastructure.
Create something you are passionate about.
If everyone did that, I think the world would be a radically better place.
Oh, and if you want to jump down the rabbit hole and discover why private payments are one of the world’s most prominent problems to solve right now, I implore you to go follow @tipz_cash.
If you enjoyed the article, you can tipz me here