GodHands, a San Francisco startup backed by Y Combinator, has built software that observes employees as they work and automatically generates training manuals for artificial intelligence agents. The company, founded by two former quantitative traders from India, posted its debut on the accelerator's Launch YC platform in October 2026.
The pitch is straightforward, if unsettling: Let GodHands watch your support reps handle customer refunds, your salespeople update CRM records, your operations team route approvals through Slack. The platform will learn those workflows, document every step with citations showing who did what and how often, then package the knowledge so AI agents can observe and report on the same tasks—though they currently require human approval to execute changes.
Whether companies actually want their employees teaching machines to do their jobs remains an open question. CEO Sai Vamshi Batchu acknowledged as much when speaking to Entrepreneur India. "Initially, I think the biggest challenge is the trust factor," he said. "People generally don't want to be replaced."
Observation Without Integration
The two-person team claims their system can learn workflows across applications like Slack, Zendesk, Gmail, Jira, and Salesforce without requiring traditional software integrations. Instead, GodHands captures text and structural data from employees' screens as they work, then assembles that information into what the company calls "playbooks."
These playbooks arrive within minutes, according to the company's website. Each line includes a citation noting which employee performed the action and how frequently the pattern appeared. An example on the site shows a refund-approval process distilled to a single rule: "Over $500 → post in #refund-approval," with attribution reading "Learned from Sam's work · 14 times."
The model feels somewhere between process mining—a category dominated by established players like Celonis and UiPath—and the newer crop of workflow documentation tools such as Scribe and Tango. But where those products focus on creating guides for human workers, GodHands aims to train AI agents that can execute the tasks themselves.
The Memory Layer
Batchu and co-founder Aditya Bommakanti both worked at Graviton, a high-frequency trading firm based in India, before launching GodHands. Batchu handled quantitative research and studied computer science at IIT Kharagpur. Bommakanti built low-latency systems and published research at NeurIPS in December 2024, the machine learning conference.
Their framing positions GodHands as complementary infrastructure to foundation model labs. "OpenAI and Anthropic are handling the intelligence layer," Batchu said. "We want to handle the memory layer for everyone."
That memory layer, in practice, means storing institutional knowledge about how work actually gets done—not how a manager thinks it should get done, but how frontline employees navigate exceptions, workarounds, and unwritten rules. The company believes this observed behavior is more reliable than manually documented processes, which tend to drift from reality.
Security Theater or Real Safeguards?
GodHands runs on-premises or inside a customer's private cloud environment, supporting AWS, Azure, and Google Cloud deployments. The platform does not capture screenshots or audio, according to the company's security documentation. Administrators can choose which applications employees are allowed to record. Password fields get emptied and keystrokes discarded locally before any data leaves the device.
AI agents trained on the resulting playbooks operate in read-only mode, the company says, meaning they can observe and report but not execute changes without human approval.
The site displays badges claiming SOC 2 and HIPAA compliance, though the company does not provide third-party certificate links or audit reports for verification. The platform supports macOS 13 and later, plus Windows 10 and up. Teams can connect OpenAI or Anthropic models, or point the system at their own endpoints.
The Business Model
GodHands joined a recent Y Combinator batch with Nicolas Dessaigne listed as the company's primary partner, according to its YC profile. Dessaigne co-founded Algolia, the search infrastructure company, before becoming a partner at the accelerator.
Pricing follows the enterprise playbook: no public rates, just a "book a demo" button. Tool directory sites list the company but provide no cost estimates, suggesting deals are negotiated case by case depending on company size and deployment scope.
The competitive landscape includes not just documentation tools and process miners, but also enterprise knowledge platforms like Guru, which offers AI-powered search across company data with source citations. What distinguishes GodHands is the focus on converting observed behavior directly into executable agent workflows, collapsing the gap between documentation and automation.
Still, Batchu's comment about trust cuts to the heart of the product's tension. Employees might tolerate systems that document their work. Training their replacements is a different proposition entirely. How companies navigate that dynamic may determine whether platforms like GodHands become standard infrastructure or cautionary tales about automation's limits.
