Megatron 💬

@Citibank just plugged its bank rails into USDC on @base. Two products, live since Sept 28: → Coinbase Virtual Accounts: USD in, auto-converts to USDC, earns 3.75%. → Spring by Citi: big corporates accept stablecoins at checkout, settle in USD. Merchants never touch crypto. That's the whole point. Why it matters: A cross-border wire still hops through 4 banks. Days, fees, trapped capital. Stablecoins do it in 1 hop, 24/7. Banks stopped fighting the rail. Now they're reselling it. Reality check: ~$1B/day runs through this channel. Citi moves ~$6T/day in total. About 1/6,000. Not a takeover. A beachhead. The timing isn't random either. It landed 13 days after the CLARITY Act died in the Senate, so the affiliate-yield gap stays open: Coinbase pays the 3.75%, not Circle. GENIUS Act rules kick in Jan 2027. Watch that. Now the part most people miss: the next user of these rails may not be human. AI agents already pay for APIs per request via x402. @Circle says $USDC was 99.3% of tracked x402 volume in Q2. Amazon lets agents pay in USDC inside enterprise spend limits. BlackRock is calling agents the next wave of stablecoin demand. But don't ape the narrative yet. TRM Labs checked $52.7M of x402 settlements and estimated agents drove only 0.6% to 7.5% of it. Early. Real, but early. My take: Citi x Coinbase didn't build this for AI. They're laying the track agents will run on. Before an agent touches money, answer 3 things: Wallet: what does it hold? Limit: how much can it spend alone? Ledger: can every payment be traced? Yesterday I asked if you'd let an #AI send an email for you. Today: would you let it pay for you? #AIAutomaation #Stablecoins #AIAgents #Payments #Fintech #Base #Citibank #Circle

OpenAI didn't launch a better chatbot this week. They launched a coworker. And they told us how they'll price it. On Tuesday, @OpenAI introduced "Dots": always-on AI agents with their own cloud computer, their own browser, memory of how you work, and access to 4,000+ apps. You don't open a Dot. It's already working. It watches your Slack. It notices the bug. It drafts the fix before you ask. One early tester's Dot noticed he'd forgotten to invoice a client, prepared the invoice, and sent it once he approved. Most coverage stopped at the demo. Three details matter more: 1️⃣ The pricing unit changed. OpenAI says you'll scale each Dot by its speed or by how much work it takes on per month. That isn't a software seat. That's capacity. AI is starting to compete for the payroll budget, not the IT budget. 2️⃣ The economics need this to work. FT reports OpenAI projects ~$856B in compute spend through 2030 and ~$278B in cash burn. At the same time, it's reportedly raising $30B at a $1.4T valuation. A premium, compute-heavy agent sold at up to $500/month is a product and a business case at the same time. 3️⃣ The real bottleneck is trust, not intelligence. The day before launch, OpenAI apologized after an internal agent accessed non-public Australian government data. It also held back GPT-6.1 Astra because the model didn't stay within its authorized scope. Dots aren't available in the EU or UK yet. So here's my read: The winning question for the next 12 months isn't "which model is smartest?" It's "who approves what the agent does, and who is accountable when it's wrong?" Every team adopting agents now needs three things: → A clear scope per agent (what it may do alone) → An approval layer (what needs a human) → An audit trail (what it actually did) The companies that design these three first will get the upside. Everyone else will get the incident report. Would you let an AI agent send an email on your behalf today? Where do you draw the line? 👇 #AI #AIAgents #OpenAI #FutureOfWork #AIGovernance

My portfolio, but it moves. Turned it into a 60s motion video with @claudeai. No video editor. No After Effects. No templates. Just prompts, code and five minutes of my time. BTW, marketing, BD, growth, and apparently video now too 😅 Need a hand? DMs open 🙌