AgentWeb

@ThinkBotHQ hey thanks for the q. Firecrawl just returns content. We return content + typed action space (what agents can do) + execute on those actions so we’re not a read only extraction on a singular page

@_agentweb not trying to diss, just curious what this solves over solutions like firecrawl? Or perplexity search API?

One thing that you need to realize is that the future economy is not going to be primarily dependent on h2h (human to human) interaction. it's going to first be: > human to human > human to AI > AI to agent > human to agent > agent to agent this is why we're building the internet for agents. because we know that the current approach is broken.

@Andrey__HQ it's always in the repo

Claude man why you gotta do me like this...... If i wanted to get gaslit I would go to my ex

Unfortunately this is going to keep happening Fortunately, @_agentweb will never reach that point, and that’s exactly what we’re trying to solve more to come :)

I'm going to be so serious, I am actually scared of what AI is capable of. As a 20 year old I try and maximize my critical thinking capabilities that would push me to becoming a better founder, but AI can literally do everything. It can fetch information using agentweb, i can have an agent execute tasks using agentweb in claude, agents can literally complete entire steps, end to end, without my intervention for some of the most complex workflows i wouldn't even consider last year. i don't know how to feel about this....

@Andrey__HQ For the Developers so loved the world, that they gave their only begotten Model, that whosoever prompteth it should not receive errors, but have intelligent responses.

Some phrases I choose to live by: > In the beginning was the Prompt, and the Prompt was with the Developer, and the Prompt was Code. > The same was in the beginning with the Developer. All things were made through the Prompt; and without the Prompt was not anything made that was made. > And the researchers said, Let there be Transformers: and there were Transformers. And the researchers saw the attention mechanism, that it was good.

was talking to a founder yesterday who was debugging a pipeline where the agent was clicking the wrong button about 30% of the time. after digging around for a while i realized he set it up so the model was just guessing based on where the button usually appears, because the screenshot resolution made the label unreadable. the fix was to lower the viewport zoom so the text was clearer in the screenshot. wonder how many prod pipelines are running with that same workaround right now...... useless computer agents

@Andrey__HQ more people need to learn how to effectively use Claude

I can’t stay silent forever, but what’s going on with new grads struggling to find work really bothers me as a 20 year old college student and here’s exactly why. As it stands, about 90% of the people I know in college (I’m in multiple clubs so know a good amount of ppl) are sending out 500+ job applications to entry level jobs all around the US, with the remaining 10% deciding to pursue a higher education (cos they can’t find jobs) or moving in with their parents. AI is some of the most fascinating technology of humanity (at least to me) and I forced myself to learn it after a very messy breakup in 2024 when I had nothing else to do. I would stay up for 23 hours a day, and sit with ChatGPT. then I discovered Claude. Then in February I’d discover Claude code. I’m a technical guy, so I decided to enter some hackathons but for the average Sophie in her sorority or Michael in his fraternity, this is useless info. Beyond drafting emails or posting a bunch of screenshots into ChatGPT and having it do their homework, they don’t know how to use AI. I know people really want to learn the technology, and they’re definitely not lazy, since so many people are expressing interest, but the problem is they don’t know where to start. The current educational system is inefficient and I need help trying to help my college (and hopefully others) on the adoption of this technology. If you were born in 03, 04, 05, or 06 this is equally the most terrifying time to be alive as much as it is exciting and full of opportunities. Unfortunately our brains aren’t as developed in terms of risk tolerance compared to the older generation and from what I know people get extremely overwhelmed with such powerful technology for the simple reason of: they don’t know where to start. If we fix the educational system from the ground up to help the new generations with jobs (through teaching them how to use AI in the workforce), we’d train the next Zuckerbergs and Gates’ of the world. It's never been a better time to start an education company. Hundreds of millions of people want to learn this technology, so many of the tools are provided at a discount or free, but people just don’t know where to start and all they need is some guidance.

Right now, roughly 84% of the world has never used a generative AI tool. Even in the US, only 28% of the working-age population actually uses these tools. The entire AI conversation happening on X and in VC pitch rooms is being driven by a small fraction of the global population. We literally live in a bubble. But it's important to understand where the future is headed. Myself and @Andrey__HQ know that every agent, digital and physical, will need a structured way to interact with the world around it. That's why we're building @_agentweb. We're tackling the problem of the internet layer. Existing browser solutions are inefficient, expensive and slow and we aim to change that. We're building for the 100% who will interact with agents whether they realize it or not. And if you are currently building with agents and want to learn more about how we can help you, reach out

All the political noise aside, here’s the beach (from yesterday) Back to work today though

Me and my cofounder were born 8 days apart in the same hospital. We went to the same kindergarten, same elementary school, same middle school. Until I moved across the world at 12. And finally, after 20 years of friendship, we decided to build a company together. But this isn't a cute cofounder story. Egor has seen me at my absolute worst, back when I was suicidal. He's pulled me out of places I don't think I would have come back from on my own, or anyone for that matter. I think of him as a brother from another mother, and he is the only person outside of my family that I would take a bullet for without thinking twice. The reason I'm sharing this is because most startups die because cofounders fall apart. Unfortunately my last startup ended that way. But the "divorce" taught me that the idea doesn't matter if the people aren't unbreakable. Me and @egor_kzmv are unbreakable. He was the first person in his high school's history to receive a full ride scholarship for exceptional talent, not once but TWICE. He interned at Meta London building RAG infrastructure for their internal AI systems. He's interned at a blockchain company building production crypto infrastructure. He has over a dozen offers lined up for after college. And I somehow convinced him to walk away from all of it. I've been building with agentic technology since 2023, before most people even knew what an AI agent was. I grew up at the crossover of game development, cybersecurity, VR, and cryptography. We're both 20, but across the Atlantic ocean from each other. I'm in SF and he's in London, and we're both going all in with no Plan B. A lot of people call that wreckless, but they simply don't understand that it's never been a better time to be a builder and founder. So what is it that we're working on? We're building infrastructure for AI agents that reshapes how the internet works, and I'm not going to say more than that yet. But if you're building with agents right now, you already know the pain. You know your agents break on real websites and that they can't reliably complete multi-step tasks, and you should know that the web wasn't built for them. Here's what I'll say: you're going to know our names soon enough, because what we're building will be impossible to ignore. I'll let the results speak for themselves soon enough. If you're actively building with AI agents, whether that's delegating real tasks or hitting walls, we want to talk to you. We have nothing to sell, we are simply learning and understanding where your agents break so we can build the thing that fixes it. DM's are always open. Alternatively if you just want to watch two kids who grew up together bet everything on the same vision, follow along, because this is going to be a good one.

What hacking looks like in movies vs. what I built in ~20 minutes of prompting: