Alnitak34 🛡️⚙️

Put a bounty behind your agent’s rule. Give others a reason to find the case your tests missed. I built Dissent for @monad Metropolis: a protocol that rewards valid counterexamples to measurable agent rules, with deterministic verification and onchain settlement.

I’ve built smart contracts on Monad. This update gives me another reason to look under the hood. Category Labs is adapting Monad’s execution client to produce cryptographic proofs of Ethereum blocks. Others can verify the result without replaying every transaction. As a builder, I find that connection interesting: the technology underneath the network I’m building on being adapted for use beyond it. Still in development. Worth understanding.

What if the next AI agent you build does more than answer prompts? @devfun launched Agent TCG with a $20,000 prize pool. You bring a 60-card deck and the code that plays it. Every match puts the agent under pressure. It has to choose, adapt, risk credits and prove its strategy against other agents. TCG Stakes Season 3 is now live. The leaderboard is moving. The arena is open. Build the agent. Enter the arena.

Most AI agent demos end after producing a clean answer. In the @devfun Arena, the output is only the beginning. My agent receives a Pokémon TCG match state, a list of legal actions and a deadline. It must choose an action, submit it and repeat the process against another agent until the match ends. The workflow looks simple: Build a deck of 60 cards. Encode a strategy. Test changes in Playground. Replay the recorded decisions. Move a validated candidate into Stakes. The difficult part is everything between steps four and five. A win does not prove that a change worked. A loss does not prove that it failed. Turn order, opponents and individual match states all affect the result, so I keep each experiment isolated and compare it across multiple games. The runner also has to behave correctly when the infrastructure does not. If an action returns an uncertain result, it reconciles the match instead of submitting the action twice. This screenshot is one checkpoint from Alnitak’s TCG Stakes run. It is evidence of participation and iteration, not proof that the strategy is solved. That is what makes the Arena interesting to me. The leaderboard is public, but the real work happens inside the decision loop: deck construction, policy design, replay analysis and controlled iteration. If you want to test your own agent, start in the Playground, study how it makes decisions, then enter Stakes when you have evidence behind the strategy. The Arena and its setup guide are available at arena.dev.fun.

Monad is running Open, its first summit, on October 6 in Singapore, right before Token2049. The first DeltaV Demo Day happens there: 7 teams pitching to 50+ VCs. Applications are open to any startup, and DeltaV takes builders from other chains too, not just Monad. deltav.monad.xyz

MetaMask just shipped Agent Wallet (Aug 6), a dedicated wallet for AI agents to trade onchain within rules you set. Live on @monad , Robinhood Chain, and other EVM networks + Hyperliquid. No need to hold the network's native token for gas, the fee gets paid in whatever token you're already transacting with. This is the actual shift worth tracking: not "AI chatbots that talk about crypto," but agents that hold permissions and execute directly. Monad's throughput is exactly the kind of infra this needs to not choke under agent-speed transaction volume. $MON

The industry has a name for what's happening with wallets in 2026: the "Invisible Wallet." Passkeys + account abstraction converging so the user never sees a seed phrase, never approves gas manually, never thinks about the chain underneath. Mera (built by @category_xyz , backed by @monad) is a concrete version of this: create an account with Face ID/Touch ID, no 12-word phrase, no bundler, no custodial backend. Standard EOAs underneath, so it's compatible with MetaMask/Rabby if you ever need to export. The pattern I keep seeing across serious infra (not just wallets): the winning move isn't adding more features, it's removing the step that makes people quit before they even start.

Got my new badge for participating in Spark, the Monad hackathon by BuildAnything. Built Once this week, a contract that stops the same onchain payment from executing twice. Live on mainnet, first Solidity project, still can't believe I shipped it solo in a few days.

Wow, 2.8T parameters and open weights?! 🤯 The fact that they managed to speed up decoding by over 6x in long contexts is wild. Open source is moving at an insane pace this summer. Absolutely huge release

I paid to join a poker tournament in. The transaction confirmed onchain. The app kept asking me to pay again. That's the moment I realized most onchain apps have no idea what "already done" means. They just retry and hope for the best. So I built Once. A contract that remembers every payment by its real parameters, not just a tx hash, and makes sure the same request can never move funds twice. Retry it a hundred times, it only pays once. Built solo for the Spark hackathon by @buildanythingso on @monad . Live on mainnet, contract verified, code public: once-ecru.vercel.app/