The Agent Angle #9: From Wall Street to Starfleet

Hey team 👋

Welcome to issue #9 of The Agent Angle.

This week we’re hopping across worlds. Wall Street agent committees, Starfleet-style clinics, and more. Agents are debating trades, copiloting doctors, stress-testing each other, and even publishing science that looks ready for peer review.

Once agents become the process instead of just a plugin, we unlock entirely new ways of working.

BTW a couple of exciting updates:

Let’s dive in.

You walk into the ER with chest pain. Instead of your doctor splitting attention between you, billing codes, and an EHR from hell, she’s fully present. The AI is doing the scut work.

That’s the play from Epic, the electronic health record giant that runs nearly half of U.S. hospitals. CEO Judy Faulkner is rolling out what she calls “healthcare intelligence”…basically a squad of AI agents baked right into the system.

The crew feels straight out of Star Trek:

Why it matters: Epic already controls the pipes, with 42% of the U.S hospital market. Startups like Abridge and Ambience have raised close to $1B building AI scribes, but if Epic undercuts them (rumored price: $80 per doctor/month, according to Avon Health CEO Maitreyee Joshi) and turns it on by default, those companies are toast.

I’ve spent enough time in healthcare to see the grind up close: doctors spend a ridiculous amount of their day drowning in admin. None of it has anything to do with actually treating patients. On top of that, medicine moves at breakneck speed, and there’s the constant pressure to keep up with the latest. No wonder doctors are burnt out.

Why don’t we let the AI help us reroute attention? The doctor can focus on care while the agent handles the paperwork and catches what humans might miss. With Cosmos doing predictive monitoring, you can flag sepsis or readmission risk before the first symptom even shows up.

That’s lives saved.

BlackRock just published a scientific research paper (shocking, I know).

They call it AlphaAgents. Instead of one algorithm making stock picks, they built a committee of AIs that argue with each other until they agree on what to buy.

The cast is straight out of an investment sitcom. One AI binge-reads 10-Ks and filings like an overcaffeinated junior analyst. Another tracks mood swings in the news and insider activity. The third lives in spreadsheets, crunching volatility and price action.

A “moderator agent” forces them into round-robin debates until they reach a consensus.

In backtests, this multi-agent fight club consistently beat solo models. Risk-neutral committees produced stronger returns, while cautious committees took smaller losses during rough patches. It is exactly how human teams work, but….without the ego

The clever bit is how they cut down hallucinations and bias. Each agent is role-prompted to stay in its lane, and their logs are auditable like meeting minutes. That’s key for institutional adoption because fund managers actually want to review why a trade was suggested instead of trusting a black box.

If the portfolio tilts toward Nvidia, you can see whether it was the fundamental reader, the sentiment watcher, or the quant who pushed it.

And this doesn’t stop at stocks. Any high-stakes decision process (compliance reviews, hiring panels, even clinical trials) could slot in its own “committee of agents.” 

So yes, your retirement account might one day be managed by a bunch of AIs arguing.

If you’ve ever done research, you know the grind: design studies, wrangle participants, crunch stats, then suffer through manuscript formatting. It can take months, and sometimes even years.

Well, what if an AI system could handle it all in hours?

Researchers from Explore Science in Brisbane just demonstrated exactly that. They built a multi-agent system that autonomously designed psychology experiments, recruited nearly 300 human participants online, analyzed data, and produced complete research manuscripts. All without human intervention.

Here’s how it worked:

In one run, the agents asked whether visual working memory and mental rotation share common limits (a decades-old debate in cognitive science). They built tasks, collected data, ran the stats, and concluded – no, they don’t. Then they followed up with new hypotheses and experiments.

Each project ran in under 17 hours and cost about $114. For comparison, a human lab team would take weeks (and tens of thousands of dollars).

We’re drowning in research papers – more than 2.8M a year, while a human might read 300 tops. I’m happy if I can get through 2 papers a week. 

Now, we might be looking at labs that never sleep, churning through hypotheses, opening discovery to places and people that never had the resources before. 

This is the first credible glimpse of self-driving science.

This is so cool it makes our brains tingle. Confluent just gave AI agents a sixth sense.

Most agents today are like bored assistants waiting for you to ask a question. They’ll answer, but they’re blind until you poke them.

….What if your agent can notice when something changes and act before you even know?

Confluent’s new Streaming Agents plugs directly into real-time data streams. Think Apache Kafka + Apache Flink, but wired for AI. Instead of waiting for human prompts, these agents are always on, reacting to live business events as they happen and taking the next step automatically.

That means an e-commerce agent that spots a competitor dropping prices and updates your store before anyone on your team even sees it. A compliance agent that jumps the second suspicious data shows up. An ops agent that catches a glitch at the sensor level and fixes it before downtime spreads.

The clever part is replayability: you can “dark launch” new agent versions against real traffic without affecting users, then A/B test them until they prove themselves. It is like training a new pilot in a simulator that runs on actual weather conditions.

This matters because most enterprise AI experiments die in prototype purgatory. The models sound smart in a demo, but once they hit production, they choke without a firehose of fresh data. Streaming Agents stitch data pipelines and AI reasoning into one loop, with governance and testability built in.

If BlackRock showed us agents can debate and Epic showed us they can care, Confluent is showing us they can listen.

AI agents are slipping into customer support, sales, and ops faster than most teams can keep up. The problem is they’re often fragile. One day your bot politely handles a refund. The next day it invents a new return policy on the fly. QA wasn’t built for that.

LambdaTest thinks the only way to test agents… is with more agents. And so they launched their Agent-to-Agent Testing platform.

Instead of letting QA engineer manually poking at a chatbot, Lambda’s system spins up a squad of 15 specialized AI testing agents – security auditors, compliance checkers, even tone critics – and sets them loose on the agent you’re trying to ship.

You feed it docs, audio, screenshots. It generates stress tests that feel brutally real: Does the bot dodge refunds? Hallucinate policies? Stay calm when insulted?

Because the system runs on Lambda’s HyperExecute cloud, they say it’s 70% faster than traditional test grids, with 5–10x deeper coverage. Translation: the kind of failures that only show up under real pressure actually get caught.

Enterprises are already throwing half-baked agents at customers. Lambda’s pitch: make them spar with a fight camp of AI testers first. If this sticks, “agent QA” could become as fundamental as unit testing.

Did you know we have a YouTube channel? We post regular research briefs. Our first one covers robotics (very fascinating area)

Not everything made our selection this week, but a few other drops worth your attention:

If this week showed us anything, it’s that agents are leaving the sandbox and stepping into boardrooms, hospitals, and even research labs.

We’re watching the early blueprints of how agents might run critical systems. That’s exciting, and a little unnerving. The question is: how long before these digital teams outperform us?

Until then, we’ll keep tracking it for you. Catch you next week ✌️

Cheers,

Teng Yan & Ayan