Open-source sales-intelligence and outreach application hosted under Sahnan’s verified GitHub account; the retained evidence does not establish sole authorship. Signal monitors buying signals, enriches companies and contacts, drafts personalized outreach, and runs email sequences through a chat-driven workspace. It uses Browserbase and Stagehand for sites without APIs and supports self-hosting with a team’s own infrastructure and model keys.
Sahnan was Fabrio’s co-founder and CEO from June 2022 to January 2025, rather than an established sole author of its software. His personal account reports six-figure annual recurring revenue and customers including Imperial College London and University College London. The retained Autodesk listing describes its Fusion 360 add-in comparing student models with expected models at successive course stages and highlighting incorrect parts inside the CAD environment.
Sahnan’s first-person post about team work at Browserbase, which supplies cloud browser infrastructure for agents and automation. He describes observing agent onboarding, discovering a broken quickstart page and invented package names, registering those names to point to the correct SDK, rewriting onboarding for agents, and changing the homepage setup call to action. He reports a working Claude integration in about 90 seconds; these are team outcomes reported by Sahnan.
Public repository pinned on Sahnan’s verified GitHub profile and hosted under his account. Its profile description identifies real-time Connect 4 battles between GPT and Claude, with live streaming and browser automation. Detailed implementation and individual contribution history are not retained.
As a director from March to October 2020, Sahnan led the society’s COVID-19 response team. His account attributes the results to a six-person team and distributed manufacturing community: more than 2,000 Slack members in three weeks, £22,000 raised, and 25,000 face shields shipped to hospitals in Cameroon, Sierra Leone, Ghana, Moldova, and Texas.
jaysahnan.com
Bio, Work & Ideas
Jay Sahnan
Conference affiliation: Browserbase
On this page
Jay Sahnan is a growth engineer at Browserbase and the former co-founder and chief executive of Fabrio, a platform for grading engineering students’ computer-aided design assignments. His work includes Agent Experience: making software easier for AI agents to discover, install, and use successfully.
From 3D printers to engineering education
Sahnan built his first 3D printer at 16 and started selling printed objects while studying aerospace engineering at university. His products ranged from costume pieces to golf ball markers, which generated £30,000 in revenue over three months. He also served as a Royal Air Force Reserve Officer in the University Air Squadron from August 2018 to August 2019, progressing through pilot training before declining a return offer to start a company.
During the pandemic, he led the National 3D Printing Society’s response team. Through distributed manufacturing, the group shipped 25,000 face shields to hospitals in Cameroon, Sierra Leone, Ghana, Moldova, and Texas. His directorship ran from March to October 2020.
From June 2022 to January 2025, Sahnan led Fabrio, turning a problem he had encountered personally into a business serving engineering educators. The company reached six-figure annual recurring revenue and secured customers including Imperial College London and University College London. Its Fusion 360 add-in gave students feedback inside their design software: at successive stages of a course, it compared their model with the expected model and highlighted incorrect parts. Students could see where their work diverged while continuing to use the same CAD environment.
Sahnan moved to San Francisco in March 2025 and joined ion Design as a contract founding engineer, building AI-powered design tools. He joined Browserbase as a growth engineer that August.
Building for agents as users
Agent Experience. Sahnan argues that onboarding must accommodate agents as users in their own right. While observing agent onboarding at Browserbase, his team found a broken quickstart URL and agents attempting to install invented SDK package names. They registered those mistaken names and redirected them to the correct package, then rewrote onboarding for agents. They also changed the homepage’s main call to action to offer agent setup. In his account of the changes, he reports getting Claude from an initial inquiry to a working integration in about 90 seconds. He calls abandonment caused by setup friction “Instant Churn”: an agent that cannot make an integration work may discard the code and try a competitor.
Signal. The open-source Signal project, hosted under Sahnan’s GitHub account, combines sales research and outreach in a chat-driven workspace. It watches for events such as hiring changes and product launches, enriches company and contact information, drafts outreach, and runs email sequences. Browserbase and Stagehand handle browser tasks on sites without suitable APIs. Teams can self-host the application using their own infrastructure and model keys, and inspect, fork, or extend its code.
Reliability measured by completed work. In his guide to cloud browser automation, Sahnan recommends separating job orchestration from browser execution, designing retries around expected failures, and capturing logs and other artifacts to debug unsuccessful runs. He recommends tracking success rate, tail latency, and cost per successful job, tying infrastructure performance to whether the automation delivers its intended result.
Jay Sahnan and Arik Bird use Ramp, BILL and Stripe to explain three separate requirements for browser agents: reaching protected websites, adapting to changing interfaces, and turning web pages into structured data.
Ramp’s receipt workflow connects a card transaction to a receipt behind a vendor login: launch a browser agent, log in, capture the receipt and match it to the purchase.
Reliable hosted browsing involves browser identity, session continuity and process behavior as well as network access. Browserbase maintains a Chromium fork to change several of those signals.
Stagehand combines deterministic browser operations with natural-language control. In the demonstrated comparison, its reduced page representation uses roughly 4,000 tokens instead of 122,000 tokens of raw HTML.
Schema-based extraction turns page content into data another program or agent can consume. The closing demo expands this research from Chipotle’s website to multiple sources and then twenty companies, while leaving financial interpretation as a separate task.
▶ Watch ↗