Soham Ganatra: Making machines that learn from mistakes

Composio is a startup that is betting on AI systems improving like human employees, and not resetting like contractors

By Himani Kothari

Last Updated: Jan 22, 2026, 11:04 IST2 min

Prefer us on Google

New

Soham Ganatra, Co-founder, Composio photo by Selvaprakash Lakshmanan for Forbes India

Join Us On

Facebook Twitter Instagram Linkedin Spotify YouTube

Soham Ganatra (29)
Co-founder, Composio

As companies start automating everyday tasks, artificial intelligence (AI) has mostly filled in as a contract worker who is quick and capable, but also forgetful the moment the task ends, unlike an employee who’s generally reliable. A US-based startup is trying to change that.

Related stories

“When you hire a contractor, you start from base zero,” says Soham Ganatra, co-founder of Composio. “An employee, over time, becomes better and you rely on them to make decisions.” Composio teaches software agents to remember mistakes, share corrections across installations and slowly accumulate practical skills that raise reliability for business use.

Founded in 2023, Composio grew out of Ganatra’s frustration with the tedium of software integrations. While working at a startup, he and other early colleagues saw developers spending a third of engineering time wiring systems together, work that rarely feels rewarding. Advances in code-generation models persuaded him that the problem was suddenly solvable. Composio built an integrations platform and added what it calls a “skill layer” on top of it: Infrastructure that lets agents interact with tools such as Salesforce and Outlook, learn from those interactions and propagate improvements across other agents using the same tools.

The result, Ganatra says, is twofold: “Real-time modifications to surrounding tool interfaces to reduce error, and a shared memory of mistakes so agents do not repeat them.” That, he argues, closes a key gap in AI models: Lack of continual learning and predictable reliability.

Click here for Forbes India 30 Under 30 2026 list

Composio’s traction is notable for a young company. Ganatra says the platform has more than 200,000 developers on it and processes over 100 million tool calls a month. The startup has raised roughly $30 million across two rounds, including a $25 million Series A led by Lightspeed in February. The company has about 25 employees and plans to hire several product and research engineers focussed on long-context and tool planning.

On competition from bigger firms, Ganatra says Composio’s narrow focus on business-critical agent infrastructure and the execution power are its defences. He adds that the firm is not profitable and does not intend to prioritise profit in the immediate future, preferring to plough returns into product development.

A serial entrepreneur with roots in IIT-Bombay and a background in building startups dealing with chatbots and risk data, Ganatra says the next year is about execution: Attracting high-profile customers, deepening technical reliability and, crucially, hiring the small cadre of engineers who can turn the idea of “human” learning into production-grade software.

Enterprises are desperately looking for agentic workflows that do work on behalf of humans, says Hemant Mohapatra, partner at Lightspeed, which invested in Composio’s Series A. “The trouble with agentic AI is that while the simpler workflows can be automated, the real value lies in stitching together complex multi-step workflows that span across multiple tools. These workflows are too brittle and have accuracies as low as less than 10 percent.”

Composio, he says, has “built the infra that, one, captures, and two, distributes practical knowledge on how these workflows work across the entire AI ecosystem”. For example, when any single agent built on Composio masters a Salesforce integration or optimises a GitHub workflow, those insights propagate to every other agent on the platform making them all better en masse. “We think this layer is critical.”

First Published: Jan 22, 2026, 11:17

Subscribe Now

(This story appears in the Jan 09, 2026 issue of Forbes India. To visit our Archives, Click here.)