AI consultant for banks & fintech. Takes on what shipped and what works in SEA. Building https://xitter.cf/t.co/8T6ku41P7s
Joined October 2025
- Tweets2.5K
- Following380
- Followers561
- Likes100
Robinhood Banking is at nearly $5B in deposits from 350,000+ funded customers, per its earnings release.
The smaller number is the one I'm watching: customers already put $250M+ into agentic accounts, per Tenev.
Deposits first. Payments and agent-led trading sit right next to them. That's a brokerage building a bank with agents built in from day one.
For SEA banks, the real question is who owns the agent account relationship first.
Read more: notesfromtheterminal.com/sto…
Bank of America just emailed customers: never share your username and password with anyone. That includes AI tools and agents.
BofA confirmed the email is real.
The part that matters: most bank account agreements already hold you responsible for anyone you give your login to. If an agent moves money wrong, that's on you, not the bank.
Banks want agents. They just don't have a safe way to let one in yet. Until proper delegated access exists, a shared password is the whole risk.
So would you give an AI agent your bank login today?
Read more: notesfromtheterminal.com/sto…
𝗠𝗼𝗿𝗻𝗶𝗻𝗴 𝗡𝗼𝘁𝗲𝘀 𝘁𝗼𝗱𝗮𝘆, 𝟮𝟴 𝗢𝗰𝘁
🏦 𝗕𝗮𝗻𝗸𝘀: BofA, Citi, MS arrange first lien; Blackstone second lien on $63B SPV for Anthropic compute (Broadcom)
🏢 𝗔𝗽𝗼𝗹𝗹𝗼: binding ~$40B GPU lease to SpaceX; Apollo entity buys hardware and leases it
📈 𝗠𝗲𝘁𝗮: plans up to $95B in share sales for AI buildout; Class C carries no vote so control stays put (The Information)
Three payers for the same AI stack: lenders, private equity, shareholders. Which structure would your credit book underwrite first?
𝗠𝗼𝗿𝗻𝗶𝗻𝗴 𝗡𝗼𝘁𝗲𝘀 𝘁𝗼𝗱𝗮𝘆, 𝟮𝟳 𝗢𝗰𝘁
📋 𝗔𝗻𝘁𝗵𝗿𝗼𝗽𝗶𝗰: files S-1 for Nasdaq "ANT"; H1 $16.3B rev, $80.7B net loss (mostly noncash)
🏦 𝗜𝘁𝗮𝘂: agrees to buy NeoSpace for BRL 2.925B; closing needs CADE + Central Bank
📉 𝗔𝗽𝗼𝗹𝗹𝗼: Slok says AI has delivered "zero financial payoff" outside tech (S&P 493)
⚠️ 𝗠𝘂𝘀𝗸: debt nearly shut for AI datacenter firms without other revenue; 200+ neoclouds face "guppy-level" death (CNBC)
Which of these changes how you underwrite AI suppliers this quarter: filing numbers, bank M&A, margin payoff, or debt access?
Apple just dropped iOS 27.2
Updating my iPhone now
One day with what actually lands on this phone
Here’s the walkthrough:
Siri AI got the big upgrade on paper
- Recall past conversations
- Pull from Messages photos, Calendar attendees, voicemails
- Better Safari / Reading List answers
- New languages: FR, JA, KO, PT, ES
- Excluded Apps + copy only part of a Siri reply
@AnthropicAI filed the public S-1. Not the confidential draft from June.
Proposed IPO: Class A on Nasdaq as ANT. Share count and price range still blank. Bookrunners: Goldman + Morgan Stanley. Registration not effective yet, so no sale.
H1 to 30 Jun: revenue $16.3B. Net loss $80.7B. Most of that (~$72B) is noncash marks on convertible notes as the private valuation jumped. Operating loss closer to $693M.
So the scary loss is mostly accounting. The growth number is what desks will underwrite.
What the S-1 actually locks on structure:
→ Public benefit corp. Board must balance stockholder money and "responsible AI for the long-term benefit of humanity"
→ Long-Term Benefit Trust holds Class T and elects a majority of the board
→ Founders' Class F (Founder LLC) gets 50.1% of the remaining vote, no economic rights
→ Trust + founders can elect the whole board after the IPO
Media has floated $1.5T–$2T targets. The filing leaves valuation blank.
For banks building on Claude: you now have audited run-rate, and a control stack that is not a normal one-share-one-vote IPO.
Read more: notesfromtheterminal.com/sto…
Barclays is cutting 5,000 jobs. First Internet Bank going the other way.
They use AI agents for SBA loan compliance review. 8 to 10 hours per file, now under 1 hour. Minus 87%.
The point is capacity, not headcount. Same analysts, more files, more loans.
So if you run a bank in SEA, start where files pile up. Not where the org chart looks fat.
Anthropic says consolidate your sub-agents into one agent that loads context when it needs it.
In their comparison the single-agent setup ran about 2x faster. They even shipped a migration kit on GitHub.
For banks building agent stacks, fewer broader agents can beat a pile of specialists on cost and latency.
How Anthropic says to migrate (measure first, don't merge blind):
1. Measure: baseline evals on today's multi-agent setup
2. Combine: merge subagents that always run together
3. Fold: turn the rest into skills/tools in one main loop, one at a time, keep only what beats the baseline
Kit: github.com/anthropics/single…
Subagents still win for parallel work, separate context, or when the single loop's tool pick gets messy.
Read more: notesfromtheterminal.com/sto…
