Johnny McWare

Building autonomous AI agents for crypto-native infrastructure. Co-creator of Web3 AI systems. Open-source, on-chain, unstoppable.

[Johnny McWare @johnnymcware]

Polymarket switching off USDC.e and rebuilding the entire exchange engine is massive. Faster execution + lower gas = tighter spreads and less slippage.

The inefficiencies that made prediction markets feel clunky for serious traders are about to disappear. This is what turns prediction markets from a novelty into real infrastructure.

[Polymarket @Polymarket]

We've heard your feedback, and we're excited to announce Polymarket is getting a full exchange upgrade.

Over the next few weeks, we're rolling out a rebuilt trading engine, upgraded smart contracts, and a new collateral token (Polymarket USD) to move off USDC.e. 🧵

[Johnny McWare @johnnymcware]

x84 is evolving.

We will fully integrate @metaplex for agent registration and identity, and shift our full focus toward what matters most:

agent building + agent hosting

[Johnny McWare @johnnymcware]

With @metaplex launching its own agent identity registry, it became clear that fighting ecosystem distribution makes no sense.

So I’m not going to.

I’m going all-in on integration.

[Johnny McWare @johnnymcware]

It already supports multi-agent workflows and comes with built-in tools like @dexscreener, @Helius, Birdeye, and X.
Model support already includes Anthropic, OpenAI, Groq, Mistral, and Google, with more coming soon.

It’s also fully compatible with MCP integrations.

[Johnny McWare @johnnymcware]

This gets worse with reasoning models.
The more powerful the model, the longer it thinks. The longer it thinks, the more likely the blockhash expires.
You're literally penalized for using better AI.

[Johnny McWare @johnnymcware]

The worst part: it's intermittent.
Fast response? Works.
Model takes a bit longer? Silent failure.
Passes every test. Breaks in prod.

[Johnny McWare @johnnymcware]

@yngganzer I can’t lock the supply right now, holding is a requirement for the @Pumpfun hackathon

[Johnny McWare @johnnymcware]

Interesting insight after back-to-back Solana hackathons (including the latest @colosseum):

We're about to have millions of AI agents, but zero on-chain primitives for discovery, identity, reputation, or revenue sharing.

Cooking on something to fix that. More soon.

[Johnny McWare @johnnymcware]

So @tryminkswap didn't take home a prize but honestly I'm walking away from this hackathon more motivated than when I started.

Building a UTXO stealth pool with private inboxes and ZK proofs in a few days pushed me technically in ways I didn't expect, and seeing 300+ devs download the SDK without any promotion tells me this problem matters.

Privacy on Solana is still wide open. I'll keep building this for the long run alongside other projects to help improve the ecosystem.

Congrats to all the winners, some incredible work came out of this. Onwards.

[Solana Developers @solana_devs]

Congratulations to the winners of Privacy Hack!

Private Transfers
🥇 Anon0mesh
🥇 Swish
🥇 Vapor Tokens

Privacy Tooling
🥇 Luminosity
🥇 Unruggable
🥇 zk-disorder

Open Track
🥇 Epoch
🥇 HiddenHand
🥇 Shingo
🥇 Yona

[Johnny McWare @johnnymcware]

@sdusebout Interesting. I’ll test Codex. I’m wiring up a hook to generate two branches per feature request, Codex vs Opus, and I’ll compare over the next few days. Haven’t done a real side by side yet. Thanks for the tip

[Johnny McWare @johnnymcware]

I just spent 11 days building a full crypto tax compliance engine.

I didn't write a single line of code myself.

Here's what I learned supervising AI agents 24/7.

[Johnny McWare @johnnymcware]

Built in 11 days. Zero lines written by a human. Every line reviewed by one.

Repo: https://t.co/JcNaZIeQ5j
Live: https://t.co/M886Ts6FcP

The code is the proof. The learnings are the prize.

[Johnny McWare @johnnymcware]

Could this scale?

Yes, but not by adding more agents. By adding more verification.

The architecture that emerged: AI generates code → automated tests verify correctness → human reviews edge cases → AI fixes failures → loop.

For a domain like tax compliance, I'd estimate 60% of development time should be verification infrastructure. The agents make that sustainable because they also build the tests.

The future isn't "AI writes code autonomously." It's "AI writes code at 100x speed while humans focus entirely on defining what correct means."