2 weeks ago, Brett Adcock posted a public browser agent challenge; the website contained 30 steps and had to be solved in under 5 minutes. The first thing that stood out to me was that I only got to step 8 in 5 minutes, a far cry from 30 minutes. Any agent that can complete this would surely have to be ‘superhuman’. In addition, no frontier model was able to solve this, so, in some sense, any solution also needs to 'push the frontier' on this particular challenge.
As it turns out, traditional post training (RL) was not the best solution to this problem instead a recursive DSPy policy pushed GPT OSS 120B from being unable to complete step 1 to finishing all 30 in 4:10 minutes (10x faster than Sonnet 4.6). Wrote a blog detailing this below.
Two freshmen, no money, no plan. Just a side project to find alumni.
Four months later, they went viral at Stanford, hit 40K DAUs, and got into YC.
Now they’re attempting to build the internet’s most powerful people-search engine, already capable of running 10,000 LLM calls at once on Groq.
This is the story of @cladoai and its mission to shake up the multi-billion dollar business of finding people online.
🧵👇
2/ Top agents to research anything
GRID brings together 100+ intelligence partners, spanning agents, models, data, and tools. Our ecosystem includes leading Web2 agents designed to support virtually any type of research.
@ExaAILabs: Real-time, citation-rich web search that retrieves the most relevant and credible sources.
@cladoai: A people-search and profiling tool that aggregates web and social media data to generate comprehensive research on individuals.
@askalphaxiv: Surfaces papers, preprints, and technical research with summaries.
Our Open Deep Search framework supplements this workflow with top-tier current information
gpt-oss prepares the answer based on all of the information provided above
@composio: Connects AI to 100+ SaaS tools, APIs, and workflows (e.g., Notion, Slack, Stripe, GitHub) so an agent take actions across apps.
all 4 of our founding engineers in one photo:
alex turned down amazon & paypal
jefferson & rohin were acquihired from jaimy
david just graduated high school
clado team is amazingly diverse 💜
In July we gave out a free t shirt to every user that did a customer discovery call with us 😤 after 74 calls in a week we are still not satisfied. If u r a clado atlas user email me at [email protected]
We’ll give u a mystery gift that will not disappoint
Ft @supportingden’s founding engineer w clado merch
Today we launched our new platform: Clado Atlas.
Atlas represents a new age of people search engine—one that understands intent, not just keywords. It works by combining small language models trained to interpret your query semantically, retrieve a precise, complete set of matching people, and return them enriched with high-quality contact data. Atlas currently outperforms all other people search engines and enrichment platforms. Benchmarks below.
But with that power comes risks and responsibility.
Clado Atlas is the most complete and accessible people search tool ever made. That requires care—in how it’s used, who uses it, and what it’s used for. We’ve built strong safeguards around enrichment, attribution, and usage. This isn’t a tool for spam or scraping. We will work hard to make sure it doesn't become one.
We're happy to answer any questions that arise surrounding safety, compliance, and privacy of data (founders [at] clado [dot] ai). We are excited to continue developing high quality products that shape the world's work.
Atlas is live now at atlas. clado. ai. Let us know what you find.