Here is everything that happened in AI Agents this week 🧵 (save for...

@OpenAI is preparing to launch a computer using AI agent codenamed “Operator” that take actions on a person’s behalf thru a browser, such as writing code or booking travel.
Thanks for sharing @shiringhaffary

Staff told in all-hands mtg today it will be released in Jan
w @rachelmetz
bloomberg.com/news/articles/…
@stripe is launched a SDK built for AI agents:
- LLMs can call payment, billing, issuing, etc APIs
- Use any model via functions (w/ per token billing)
Great work @jeff_weinstein + @stevekaliski
@braelyn_ai 🚢 the AgentStack <> @stripe integration

- LLMs can call payment, billing, issuing, etc APIs
- Integrates with @vercel, @langchain, @crewAIInc
- Use any model via functions (w/ per token billing)
Excited what you, and your bots, build: stripe.dev/blog/adding-pa…
@withneo launched the first Autonomous Machine Learning Engineer.
NEO is a multi-agent system that automates the entire ML workflow, saving engineers thousands of hours of grunt work.
Excited to test this out!

NEO is a multi-agent system that automates the entire ML workflow, saving engineers thousands of hours of grunt work.
Our recent advancements in multi-step reasoning and memory orchestration have enabled NEO to excel at solving complex Machine Learning problems on its own from data engineering to deployments of ML models.
Benchmarks:
NEO was tested against 50 Kaggle competitions and scored a medal in 26% of them, far superior than the previous state of the art performance of 16.9% by Open AI’s O1 with AIDE scaffolding on the MLE bench
With these capabilities, NEO aims to help every ML engineer become superhuman and focus on real innovation instead of the existing grunt work
Check out NEO in the thread below:
@braelyn_ai released a new version of AgentStack.
- remove tools with CLI (@ssslomp)
- @stripe agent tooling (thx @jeff_weinstein @stevekaliski)
- @perplexity_ai agentic search
- install via Brew
- @ExaAILabs agentic search
 bug fixes

↩️ remove tools with CLI (@ssslomp)
💵 @stripe agent tooling (thx @jeff_weinstein @stevekaliski)
🧠 @perplexity_ai agentic search
🍺 install via Brew
🔍 @ExaAILabs agentic search
🪳 bug fixes
hiring?? 👀
Official statement from @pyautogen about AutoGen forks and clones and the current pyautogen package.
Very interesting.
The awesome team @firecrawl built an OSS daily AI newsletter that's completely generated by web connected LLM agents.
Agents crawl the web for news, generate a newsletter, then send it right to your inbox!
Great work @CalebPeffer @ericciarla

A daily AI newsletter that's completely generated by web connected LLM agents.
Agents crawl the web for news, generate a newsletter, then send it right to your inbox!
Open source & powered by @OpenAI's o1, @resend, and @firecrawl.
@CMD_LABS shared a diagram showing how @AnthropicAI’s Claude Computer Use works.
He gives a two implementations that anyone can use.

Definitely a lot of privacy and data security implications introduced with this design, but it's hard to deny the innovation.
A+ on innovation
F- on personal data security (which @AnthropicAI would agree with so I'm not hating).
This short blog highlights some of the data security concerns: anthropic.com/news/developin…
Here are links to 2 amazing implementations of Claude Computer Use if you want to give it a spin...
IMPLEMENTATION 1 - Original @AnthropicAI example
LINK TO CODE: github.com/anthropics/ant…
IMPLEMENTATION 2 - @AgentOpsAI DX-enhanced example
LINK TO CODE: github.com/AgentOps-AI/do… (fork authored by @teocns)
Shout out to @prattyboy, @AlexReibman, @adamsilverman, @AgentOpsAI, @braelyn_ai, @DarioAmodei, @DanielaAmodei
#anthropic #agentops #docker #streamlit #claude #vscode #vnc #novnc #containers #agents #ai #ml #safety #datasecurity #innovation
@bhancock_ai shared a new tutorial on building @crewAIInc flows.

🔁 CrewAI Flows: AI Feedback Loop 🔁
Brandon covers how to set up a flow that evaluates its outputs, iterates with feedback, and polishes each result until it meets your standards!
Watch the full tutorial here: bit.ly/3UOozSK
@virattt made a diagram of his AI Hedge fund.
It has 4 agents:
• portfolio manager
• fundamental analyst
• technical analyst
• sentiment analyst

We have 4 agents:
• portfolio manager
• fundamental analyst
• technical analyst
• sentiment analyst
We use LangGraph to decide which analysts to run.
Portfolio manager delegates work to analysts and summarizes their research.
Query #1: “What is the latest net income + revenue for AAPL?
The PM delegates this task to the fundamental analyst.
Query #2: “What is the latest price action and news for AAPL?
The PM delegates this to the technical and sentiment analysts.
Eventually we will have multiple teams that compete with each other.
The agents have access to data for 30,000+ tickers using @findatasets
@hwchase17 shared the biggest blocker for AI agents is just getting them to work.

It is performance quality, BY FAR
For all the talk of cost, latency, safety... the fact is, most people are still just struggling to get agents to work 🤷♀️
Full survey here: langchain.com/stateofaiagents
@SullyOmarr shared that the most impressive thing about research agents is how
1) versatile they are
2) they remove the learning curve (stuff that took months to learn, anyone can do instantly)
3) its waaay faster than a human

1) versatile they are
2) they remove the learning curve (stuff that took months to learn, anyone can do instantly)
3) its waaay faster than a human
for ex, if you're looking at etherscan wallets its really hard to tell whats legit or not. but with the right research agent workflow you can do it almost instantly
I shared my perspective that research agents will be one of the biggest use cases over the next year.

It is physically impossible for a human to access 10,000 websites in an hour and summarize the data. However, an agent can do this with ease.
We have gotten 10+ requests in the past week.
@yersultansapar built an agent that creates a test plan for Google Maps, executes it, and files a bug to Jira.

See how our agent creates a test plan for Google Maps, executes it, and files a bug to Jira!
#Commerce x #AIAgents = the ‘Agentic Commerce Era’
@BainCapVC's Scott Friend coins the phrase to signal we’re at the beginning of a significant wave of innovation in commerce marked by intelligent agents working on shoppers’ behalf.

@BainCapVC's Scott Friend coins the phrase to signal we’re at the beginning of a significant wave of innovation in commerce marked by intelligent agents working on shoppers’ behalf.
A sampling of portfolio companies poised to power progress into 2025 (and beyond): @attentiveHQ, @bloomreach_tm, @ShipBob, @Mirakl, @archiveresale, @soonastudios, @fermatcommerce.
Overview below. Scott’s full, in-depth perspective here: baincapitalventures.com/insight/the-da…
@sarahzorah shared how AI agents will become the primary software users in workplaces. Humans will do what we have always done: have conversations.

Seeing more companies lean into chat-based interfaces, enabled by AI agents. People prefer conversational communication. This is why a huge amount of knowledge work happens in Slack and Teams, where people spend hours of their day.
The future will be collaborating with AI agents to do performance reviews, interview notes, project updates and having it search from and update the company’s tools and docs.
@KuraAIAgents
(YC S24) just built the new state-of-the-art for Browser Agents. It scored a world-first 87% on the WebVoyager benchmark, beating Claude's Computer Use demo by 31% and the previous SOTA by 14%.
@xRonitBasux+ @hwangdarren

ycombinator.com/launches/MGd-n…
Congrats on the launch, @xRonitBasux + @hwangdarren!
@amasad gave a great talk on @Replit agent
Thanks for sharing @grinich

also reminds me of the golden era of meetups when i first came to silicon valley. super ambitious people pushing cutting-edge tech. excited for more of this @amasad
Solid agent idea:
“Everyone will scream “Agents, agents, agents!” but that can come later.
… I should be able to have a dozen tabs open, @ a specific tab or site, and it should pull in information.
Would be a hit.” - @mckaywrigley

Build a browser with AI chat & context tools - similar to Cursor.
Example: I should be able to have a dozen tabs open, @ a specific tab or site, and it should pull in information.
Would be a hit.
We are actively hiring at @AgentOpsAI for our agent evals team at AgentStack team. DMs are open.

Looking for 10x engineers who are interested in building out the most comprehensive agent evals tool.
DMs are open.
Please share with anyone in your network who is interested.
“tremendous pressure on every CIO / CTO to come up with some AI success stories so that the company isn't seen as missing out on AI, so of course they have responded by rebranding traditional automation as "AI agents". - @random_walker

AI Engineering Agent for GitHub can fix bugs and create pull requests.
Just connect your GitHub repo, ask LLM to generate code or fix bugs, and even create a PR.

Just connect your GitHub repo, ask LLM to generate code or fix bugs, and even create a PR.
@svpino shared the complete source code of an AI agent that can analyze tabular data.

If you want to learn how to build agentic workflows, check this out because you'll learn a ton from this.
This agentic workflow will do the following:
• Configure the agent with tools for accessing external APIs
• Define a goal to encourage the agent to generate and run code
• Track the agent's reasoning process
The code is written in Python, and it uses Dynamiq, an open-source orchestration framework for agentic AI applications.
Here is the repository link: github.com/dynamiq-ai/dyn…
The best way to learn is to get your hands dirty!
Meet your new AI executive assistant.
@floodeai founder
@SarahAllali7 shows how Floode works to automate your email with @Scobleizer
x.com/Scobleizer/sta…
Did I miss anything? Share below.
Liked this thread? Share it with your friends/team.
Follow @adamsilverman and @AgentOpsAI for everything AI agent related.
Last weeks thread:
x.com/AtomSilverman/…










