BLOG / ai agents
Nobody hires an assistant to manage it
Assistant Benchmark lists 132 AI assistants and four for finance. The a16z episode on them made the case for the kind that reports back on its own.
There's a line in a16z's September 29 episode on personal agents that I'd put on the wall of anyone building one. David Pawlan, who runs Assistant Benchmark, was describing the best employees he'd ever had: they're "just doing their work, getting things done, they update you like, here's what I did, and you're like, wow, amazing." Then the punchline: "you don't hire someone because you enjoy managing them."
Anish Acharya, the a16z partner hosting, spent 50 minutes pushing on what makes an agent worth paying for. Pawlan's answer kept coming back to that picture: an assistant you mostly don't see, that shows up with a report.
I build Keepfolio, a macOS app for people who run their own brokerage account. That picture is the product brief I wish I'd written.
A crowded field that still waits to be asked
Pawlan's site is a public scorecard for AI assistants you can text. As of its September 28 update it lists 132 of them and has tested 27 on the same 15 tasks, scored 1 to 10 after real use. Four sit in the finance category.
A chat assistant waits for a question. Pawlan thinks the winners won't: "I think there is massive defensibility around proactivity. I think that's actually going to be something that separates the winner from the meat of the pack." The best moment so far, he said, is an agent checking you into your flight.
He also named when help matters most: "an assistant is most helpful when you are most occupied."
What "here's what I did" looks like for an investor
For someone running their own account, the thing that gets skipped is the morning read. Did anything move overnight that matters to the plan, and is margin creeping toward the ceiling you set?
Keepfolio's answer is a briefing that's written before you sit down. Cassandra, the analyst inside the app, checks what moved overnight, measures it against the strategy contract you set, and writes one page: what changed, why it matters, and the evidence behind each line. You don't type a prompt to get it. The facts trace back to the filings, releases, and events they came from, so you can check any line instead of taking it on faith.
That's Pawlan's employee who updates you. The difference from a chat assistant is who starts the conversation. You don't have to know which question to ask on a given morning, because the briefing already asked it.
Where the briefing stops
Pawlan was careful about where proactivity stops. An agent that finds you a cheaper insurance plan "should probably get your permission before it actually makes that switch," and "if you cross that line once, I think you immediately lose all trust with your user." Acharya had set that up a minute earlier: a finance agent is one where "you want it to be highly cautious."
Keepfolio splits it the same way. The brokerage connection is read-only, and there is no order path or money movement anywhere in the app. The briefing surfaces what to look at. You decide what to do, and you do it in your brokerage.
The case against
Two objections came up in the episode, and both are fair. Ben Thompson, as Acharya relayed him, thinks most consumers are "looking to spend time, not save time," so a daily report reads like homework. And the invisible agents Pawlan praised are invisible because they act: they claim the credit and tell you afterward. A briefing still has to be read.
I'd answer the second one first. For a brokerage account, the report is the part that should be visible, because the decisions it feeds are yours to make and hard to undo. What should disappear is the digging behind it. As for Thompson, he's describing most people, and most people don't run their own margin account. The operator who does already spends that time each morning. The briefing is the same read, done before they sit down.
Cost is the other half
Near the end came a number for doing this ambitiously: "by our estimate, it's costing something like $20 per user per day." That's the economics of a general agent working across the web on your behalf. Keepfolio's job is narrower on purpose: your own accounts, one briefing each morning, and reasoning that runs on the AI provider you choose with your own key.
SOURCES
- youtu.bea16z: We Tested the New Wave of Personal AI Agents
- assistantbenchmark.comAssistant Benchmark
Keepfolio reads a brokerage account through a read-only connection and never places orders. The morning briefing files itself