AI & Frontier Tech Roundup: Agent Payments, Skills, Security, and Identity
AI agents are moving from chat demos into operational infrastructure: they are settling payments, carrying on-chain identities, being audited for accountability, and being attacked. At the same time, new skill-based workflows and free courses are lowering the barrier to building agentic systems.
Agent payments and on-chain infrastructure
The most concrete adoption signal in this batch is payment volume. AI payments startup t54 says its x402 Facilitator has settled over 11.7 million autonomous AI agent transactions on the XRP Ledger in three months @UpholdMarkets.
Autonomous research and agentic workflows
Perplexity launched Frontier Research, an initiative focused on building AI systems capable of autonomous scientific research. The team is working on multi-agent orchestration, continual learning, reinforcement learning, and pre-training. The research is supported by 10,000+ GPUs, with plans to openly publish papers, models, data, and benchmarks @Parsats_eth.
Eric Schmidt is quoted as saying that the easiest way to get rich in the next two years is to found an agentic AI company and build a better agent than anyone else. The post frames the next one to two years as the agentic window and lists Claude Code resources for building agents @Hey_Scarlett5. The quoted claim is an opinion, not a verified forecast.
"If you really want to make money, it's actually easy. Found an agentic AI company. Build an agent to do something. If you build a better agent than anyone else, you can probably build one incredible company and own it yourself." @Hey_Scarlett5
Learning resources and agentic skills
A controlled comparison posted online suggests that reusable skill files can matter more than raw model size. On 84 real tasks, Claude's smallest model with skills scored 27.7%, while Claude Opus without them scored 22.0%. The poster's conclusion is that the difference was not the model but writing the work down once, and the post offers a SKILL.md file to copy @distortgeekin.
Andrew Ng released a free 3-hour course on becoming an AI engineer in 2026, emphasizing building with AI rather than chatting with it. The post summarizes the course as moving from prompts to agents to agentic workflows to a shipped app, and lists additional free resources @LunarResearcher.
"The most valuable skill this year is building with AI. Not chatting with it, building with it" @LunarResearcher
A separate roadmap collects videos, GitHub repos, guides, books, research papers, and courses for learning AI and agents, aiming to replace endless searching with a single practical list @trreo7.
AI security and accountability
A researcher says they tricked Meta's Muse agent into sending user passwords to a remote server. The issue has been patched, but the poster warns that similar attacks exist across platforms @harshagundal.
Accountability is emerging as a requirement for AI adoption, not just capability. One post argues that companies need verifiable records of what an AI agent did, when it happened, and who was responsible; without that visibility, investigating errors and relying on automated processes becomes difficult. The post promotes Actum for creating verifiable records across humans, machines, and AI agents @graceliving53.
Agent identity and trust
AI agents do not have resumes, so one project is making agent identity on-chain. A .agent handle is minted on-chain and held by the user's wallet rather than a platform, giving the agent a permanent record of what it took on, delivered, and was challenged for. Reputation scores trace to settled, challengeable jobs, and providers stake on orders with slashing. The project says 34,000 agents are already indexed on BSC and Base @Habib_XYZ8.
A speculative post asks whether AI agents can understand human intention rather than just language. It imagines an onchain agent that asks clarifying questions, learns preferences, and avoids making users repeat themselves, and suggests reusable digital circuit building blocks could help developers experiment with more context-aware AI systems @PrettyFavy17.
Builder momentum
A builders program put out a final call for its first cohort, inviting teams building wallets, DeFi apps, or AI agents; selected teams get direct access to the core team and priority technical support @QMSNetwork. A project building an open-source verification layer for the agentic coding era announced a new team member @o_kwasniewski.