Near’s crosschain AI agent can launch memecoins and will soon book your flights

Most AI assistants can draft an email or summarize a document, but very few can actually touch money. Near Protocol wants to change that by handing AI agents the keys to crypto wallets and pointing them toward both Web3 and the real world.

At the Redacted conference in Bangkok, Near co-founder Illia Polusukhin—one of the authors of the original Transformer paper—unveiled the alpha release of the NEAR.AI Assistant. The pitch is straightforward: an autonomous agent that handles crosschain crypto maneuvers today, with an eye on taking over everyday Web2 chores tomorrow.

From memecoin generator to flight booker

During a live stage demo, Polusukhin showed how quickly the assistant moves. With a simple prompt, the AI spun up a frog-themed memecoin from scratch: writing its backstory, generating a logo, designing a mock site, and deploying it instantly on Near's version of Pump.fun.

It also searched the web for custom T-shirt printers and generated an interactive storefront mock-up on the fly.

Booking travel and ordering takeout, however, are not quite ready for prime time. Polusukhin admitted that flight booking remains a prototype rather than a live feature in the current alpha. But he stressed that the platform is open, meaning third-party developers can build and plug in custom task agents that instantly become available across the entire user base.

Chain abstraction and agent-to-agent haggling

The real engine behind the assistant is chain abstraction, packaged as Near Intents. Instead of wrestling with bridging protocols or intermediary tokens like THORChain's RUNE, the system relies on chain signatures that allow smart contracts to swap native assets directly across Bitcoin, Ethereum, Arbitrum, Base, and Near.

Aurora Labs founder Alex Shevchenko demonstrated this by connecting a MetaMask wallet, trading USDT for USDC, and converting that into native Bitcoin. Support for Solana and TON is slated before year-end, with Dogecoin and XRP down the road.

Even more interesting is the roadmap for agent-to-agent interaction. Shevchenko explained that if your personal AI doesn't know how to set up a specific account, it can hire another specialized AI agent on the network, agree on pricing and dispute resolution terms, and settle the payment autonomously. Fiat ramps for Web2 purchases will be handled in the background by market makers like Peanut Trade tapping into centralized liquidity.

Why it matters

Giving AI direct access to multichain liquidity makes a ton of sense. Navigating bridges, slippage, and distinct network gas fees remains crypto's biggest usability nightmare. Near's approach—tucking all that plumbing behind a natural-language intent layer—is clever and genuinely useful for decentralized finance power users.

Still, we think handing operational wallet keys to an alpha-stage AI model carries massive risk. Prompt injections, logic hallucinations, and smart contract edge cases mean fund safety is anything but guaranteed.

Moreover, the leap from Web3 swaps to real-world tasks like booking flights is much harder than it sounds. Traditional Web2 platforms guard their checkout flows with intense bot detection, captcha puzzles, and strict KYC verification. An autonomous agent negotiating fiat conversions via market makers to buy airline tickets will run into regulatory and platform-level friction almost immediately. It is an impressive proof of concept, but letting it hold your travel budget right now would be premature.