Human Behavior launches agents that turn session replays into product fixes

Why it matters

Human Behavior is trying to own the step after analytics: acting on user evidence. Success would pressure dashboard vendors to become supervised execution systems.

Human Behavior launched a new version of its product on August 10th that uses AI agents to turn evidence from user sessions into proposed code changes and customer outreach, extending the startup's original pitch beyond analyzing product usage.

https://x.com/OfficialAmogh/status/2086845030568743030

Amogh Chaturvedi (@OfficialAmogh), Human Behavior's co-founder and CEO, announced Human Behavior v2 in a thread on X. His description was blunt: traditional product analytics explained what happened, while Human Behavior would now "just handle it."

Human Behavior says the new system records session replays, clicks, page views, funnels, console logs, JavaScript errors and network calls. Agents then monitor those signals for problems such as dead clicks, failed checkout flows and customers showing signs of churn. A detected software bug can become a proposed pull request, while an at-risk customer can trigger a drafted message.

The product is designed around an approval step. Human Behavior says patches run in a sandbox, each pull request includes the relevant session replay, and the agents do not automatically merge code. Those constraints matter because Human Behavior is asking product and engineering leaders to give an analytics vendor access to both sensitive user activity and the code used to respond to it.

From watching sessions to changing the product

Human Behavior originally used vision models to process session-replay video and answer questions that event-based analytics often miss. Instead of requiring engineers to manually instrument every interaction, Human Behavior said its models could watch users move through an application and identify patterns behind conversion, retention and churn.

Human Behavior v2 adds an execution layer to that analysis. Human Behavior describes a repeated loop in which agents watch sessions, identify problems, prepare an action and resume monitoring. The available actions include opening pull requests, drafting responses, researching accounts, writing product reports and preparing data queries.

Users can also ask questions in plain English through the dashboard, Slack, email, SMS or WhatsApp. Human Behavior says answers include the session replays and data queries used as evidence, giving product managers and engineers a way to inspect an agent's conclusion before acting on it.

Installation starts with the command npx @humanbehavior/wizard@latest, which Human Behavior says can connect the system to a project in about five minutes. The website lists support for React, Next.js, Vue, Angular, Svelte, Nuxt, Remix, Astro, Gatsby and standard HTML applications.

The move puts Human Behavior into a wider fight over where analytics software ends and agentic software begins. Mixpanel, Amplitude and PostHog built businesses around collecting and organizing product events. Session-replay tools added visual evidence. Human Behavior is betting that the valuable layer will be the software that connects those records to an approved intervention.

A founder-led pivot from accounting to behavior

Chaturvedi founded Human Behavior with Skyler Ji (@skyguy_skyguy) and Chirag Kawediya. Chaturvedi met Ji and Kawediya at a hacker house he organized in 2023 after his freshman year at Stanford, according to a September 3rd, 2025 TechCrunch profile.

The founders first built Dough, an accounting product for e-commerce businesses. They sold Dough to Employer.com for six figures after entering Y Combinator and pivoted into Human Behavior. Chaturvedi told TechCrunch that Dough's customers could see which products were selling, but still needed an explanation for why shoppers behaved as they did.

That question shaped Human Behavior's first product. The founders brought experience from computer vision research at Stanford's NLP group and the University of Chicago's database group, according to Human Behavior's Y Combinator profile. Chaturvedi had also studied computer science at Stanford and previously worked with Founders Fund and cloud infrastructure startup Ubicloud.

Human Behavior's current launch preserves that original focus on behavioral evidence while widening the product's authority. The founders are effectively betting that session data can supply enough context for agents to make useful decisions across engineering, product and customer operations.

The $5 million round dates to 2025

The $5 million Chaturvedi cited in the launch thread dates to Human Behavior's seed financing announced in September 2025. The backers were General Catalyst, Paul Graham (@paulg), Vercel and Y Combinator.

TechCrunch reported at the time that Human Behavior closed the round in two days. The financing followed the Dough sale and gave the founders capital to build Human Behavior around a broader premise: analytics software should produce an action rather than leave another dashboard for employees to interpret.

Human Behavior v2 is the clearest expression of that premise so far. The technical test will be whether its agents can trace a user problem to the correct cause and propose a safe response consistently. Its stated safeguards keep humans responsible for the final action. The commercial test is whether that supervised workflow saves enough time to replace separate analytics, replay, error-monitoring and product-operations tools.