The TRXN Postmortem: Why I’m Shelving a Product I Spent Years Building

July 18, 2026

I spent years building TRXN, a system that verifies cryptocurrency payments against invoices. Price in fiat, collect in crypto, verify on-chain. Multi-invoice allocation, phase-based subscriptions, payment claims, an x402 gateway for AI agent payments. It works. It’s in production. It’s good software.

The idea goes back to 2021. It started as a component inside another product I was building, one piece of a bigger system. In 2022 I pulled it out into its own side project. Over 2023 and 2024 it grew into a product of its own, and that product was TRXN.

And I’m shelving it.

The code didn’t fail. The thesis did. This is me writing down why, so two years convert into a documented lesson instead of a quiet regret.

The original thesis

All money eventually moves on-chain. People get comfortable holding Bitcoin and Ethereum, then start spending it. Businesses need to accept it, price in fiat, and reconcile what shows up on-chain against invoices. Small TAM now, huge later, and I’d be early.

I was half right. Money is moving on-chain: B2B stablecoin payments grew 733% year over year to roughly $226 billion. Visa, Mastercard, and Ripple joined the x402 standard. Stripe paid $1.1 billion for Bridge. Cloudflare is building payments into its edge network.

The rails thesis was correct. The asset thesis was dead on arrival, and I didn’t see it for two years.

The tax rule that kills volatile-asset payments

The IRS treats cryptocurrency as property, not currency. That’s IRS Notice 2014-21, reaffirmed in the digital assets guidance and the virtual currency FAQ. Every time you spend crypto, you’ve disposed of property, and you owe capital gains on the appreciation.

Buy $1,000 of gear with Bitcoin you bought cheap: you pay sales tax like everyone else, plus capital gains on the Bitcoin’s rise, plus a line on your tax return. Every transaction.

In other words: paying with Bitcoin is voluntarily choosing the most expensive way to buy anything.

Stack Gresham’s law on top: if you believe your BTC is going up, you spend dollars and hoard the BTC. The more bullish people are on an asset, the less they’ll ever pay with it. That’s why Bitcoin became digital gold instead of digital cash, and why stablecoins ate payments in three years. No appreciation means no gain to tax and nothing to hoard. Payments want boring.

My “graduation” thesis said people would move from stablecoins up to BTC and ETH. The tax code says the migration runs the other way, permanently. Everything I built for volatile assets was machinery for a behavior the IRS makes economically irrational. That machinery was my differentiator.

The rails belong to the giants

Fine, stablecoin world. What about my x402 gateway for agent payments?

As of July 2026: Cloudflare announced x402 enforcement at the edge for its millions of customers. Stripe shipped x402 support in February. Circle runs batched settlement down to fractions of a cent, plus the marketplace where agents find sellers. x402 will be owned by whoever owns the rails. A standalone gateway from a solo founder loses on distribution, price floor, latency, and discovery at once.

And the prize isn’t worth the fight yet: x402 moved about $24 million in the last 30 days, total, down 92% from the December peak. The giants can wait a decade for that to mature. I can’t.

What’s left isn’t a business

Strip those away and what survives is stablecoin accounts receivable: verify USDC against invoices, allocate, handle subscriptions, sync the books. Real workflow. Not a company. Three reasons:

The buyer is the worst buyer in software. Small crypto-native companies are maximally technical, allergic to paying for tools, and armed with AI agents. I watched a CFO vibe-code his entire revenue operations app with Claude Code. When your target customer can replicate your product in days and enjoys doing it, you don’t have a moat problem. You have a market problem.

Stablecoins deleted the hard part. A $500 invoice paid in USDC is $500, exact match. What’s left to reconcile is real but thin, and thin problems get absorbed as features by whoever owns the flow.

The TAM math caps out. Maybe 2,000 realistic customers at $10K a year is a ~$20M ARR ceiling, of which a winner takes a fraction. A nice small business. Not what I was building toward.

The graveyard I should have studied first

Why has nobody built this? haunted me the whole time. I treated the question as an opportunity. It was a warning, because the answer is: they did. BitPay since 2011. Coinbase Commerce. Request Finance. Loop. Copperx. Fifteen years of funded, competent teams attacking “businesses accepting crypto.”

Not one breakout.

That’s not fifteen years of bad execution. That’s the market repeating itself: the tooling was never the bottleneck. Demand was. When an “obvious” gap survives a decade of smart, funded attempts, the gap is usually not a gap. It’s a grave.

The product is shelvable. The thinking patterns follow you to the next thing. So here they are.

1. I mistook missing tooling for missing demand. No great crypto AR tooling looked like opportunity. In a 15-year-old market, it was evidence of absent demand.

2. I knew 90% of the tax rule. The missing 10% killed the thesis. I’ve been in crypto for years. I knew selling triggered capital gains. What I never connected: the IRS treats spending as selling. My whole model needed people to pay with appreciating assets, and that one connection makes doing so irrational.

The trap was partial knowledge. If I’d known nothing about crypto taxes, I’d have asked a professional. Because I knew most of the rule, I believed the tax picture was handled and stopped looking. Knowing 90% of a rule is more dangerous than knowing none of it, because the 90% stops you from looking for the rest. For what it’s worth, this is the most common tax misconception in crypto: exchanges report sales, nothing reports purchases, no wallet tells you a taxable event just happened. Still my miss. The fix: when your thesis touches money movement, buy an hour with a crypto CPA before you spend a year on the build.

3. I felt like I was missing something for two years. I was. I researched the nagging question repeatedly and came up empty, so I settled on “small space, few people looking.” The feeling never left, and it was right. The answer hid behind the rule I thought I already understood, so no search I knew how to run could reach it. It took AI models connecting a payments question to tax law before it snapped together. A question that survives years of searching isn’t unanswered. The answer lives outside where you’re looking. Change who and what you ask.

4. I knew payments wanted boring money. I bet on the wrong way to get there. Volatility is exactly why I built slippage margins: absorb the price movement so the money behaves boring even when the asset isn’t. I engineered the boringness in. The market swapped to an asset that’s born boring. So when USDC took off, I read it as one more asset gaining share, not the market solving volatility by substitution. You can’t engineer a property into an asset when the market can get that property by swapping the asset. I loved crypto for the store-of-value story and rails that beat banking’s hours, so I read the market through my own attraction instead of the job businesses needed done.

5. AI changed what’s worth building, and I only noticed what it did for me. I built the first three quarters of TRXN by hand. Last September I started building with agents, and the final quarter came together faster than any stretch before it. I took that as my advantage. I never flipped it around: the same agents sit in my customers’ hands, and any product whose value is “this is complex to build” got repriced. I was using the new tools while defending a moat the new tools had drained.

6. I collected validation instead of evidence. People online said “that would be useful.” Nobody pre-paid. “Would you use this” is a compliment, not a purchase order. I knew that rule. I let it slide because compliments felt like progress.

What happens to TRXN now

Shelved, not torched.

TRXN stays alive as proof of work: a production-grade, multi-tenant payments platform with on-chain verification, subscription billing, a compliance program, and a full API. Three quarters built by hand, the last quarter with agents. As a portfolio piece, it says more about what I can build than any resume line.

It also holds options that cost nothing to keep: if agentic payments mature into real volume, the reconciliation problem I built for arrives at scale and the shelf opens. If everyone vibe-codes their own AR front end, the hard 20% underneath could sell as an API. And a working platform is sellable any day I decide being done is worth five figures.

Zero maintenance ambitions. Zero evenings owed.

The lesson

The market doesn’t care how good your machinery is if the behavior it depends on never arrives. Verify the behavior first. Study the graveyard before you break ground next to it. Answer your nagging question on paper before it costs you years.

Two years, one production payments platform, and a permanent upgrade to how I evaluate ideas.

I’ve had worse trades.

-Donn