Anthropic just launched Claude Managed Agents — AI agents that run autonomously in the cloud. You define a task, they handle the rest. (Full docs here)

I wanted to see if it actually works. So I tested it.

The Test

I gave the agent my stock portfolio and one instruction: "Brief me like a personal financial advisor over coffee."

Three API calls to set up (create agent, create cloud environment, start session). Then I sent this prompt:

Here is my investment portfolio:
- AAPL (Apple) — 50 shares
- TSM (TSMC) — 30 shares
- NVDA (Nvidia) — 25 shares
- GOOG (Alphabet) — 20 shares
- LUV (Southwest Airlines) — 100 shares
- BA (Boeing) — 15 shares
- BTC (Bitcoin) — 0.5 BTC

For each: current price, weekly change, major news,
and your recommendation (hold, buy more, or take profits).

Give me total portfolio value, best/worst performer,
and your #1 action item for me this week.

Write it like a financial advisor briefing me over coffee.

Hit enter. Watched.

It didn't ask questions. It just started searching — 7 searches for prices and news, then 4 more for analyst ratings and earnings dates. Then it wrote the entire briefing. 11 searches → 1 comprehensive output → done.

The Results

Here's the actual output the agent produced. This is a direct representation of what came back — formatted for readability:

What Surprised Me

• It caught same-day news. The US-Iran ceasefire happened that day. The agent found it and factored it into Boeing and Bitcoin analysis.
• The Southwest insight was non-obvious. It didn't just say "airlines are down." It found that Southwest specifically ended its fuel hedging program — a detail that required connecting multiple data points.
• It made strategic calls, not data dumps. "Sell LUV and buy GOOG" is a specific portfolio rebalancing recommendation with reasoning.
• The concentration risk flag. Without being asked, it calculated that Bitcoin was 46% of the total portfolio and called it out.
• The tone was right. The "advisor over coffee" instruction made a massive difference. It felt personal, not robotic.

The Verdict

Three things surprised me:

• It caught same-day news. The US-Iran ceasefire happened that morning. The agent found it and factored it into the Boeing and Bitcoin analysis without being told.
• The Southwest insight was non-obvious. It didn't just say "airlines are down." It found that Southwest specifically ended its fuel hedging program before oil spiked — connecting multiple data points to explain why.
• It made real calls, not safe ones. "Sell LUV and buy GOOG" is a specific portfolio rebalancing recommendation. Most AI outputs would give you a disclaimer and say "consult your advisor."

Three minutes. Three cents. Real data, real news, real recommendations.

It's not perfect — there's no scheduling, no persistent memory between sessions, and it runs on API credits not your subscription. But as a first test? This actually works.

What's Next

So... the agent can research stocks, pull real-time prices, analyze news, and give me specific buy/sell recommendations.

You know what would be really interesting? Connecting it to the E*Trade API so it can actually execute trades. Give it my portfolio, let it continuously monitor performance and breaking news, and auto-trade based on its own analysis. A fully autonomous AI hedge fund manager running 24/7 in the cloud for three cents a day.

I'm not saying I'm going to do that.

I'm saying the API exists, the agent works, and the only thing between here and there is about 20 lines of code and a concerning lack of self-preservation instinct. 😏

Stay tuned.