From Failed Pivot to $30M Series A: Composio's AI Agent Growth

Soham Ganatra's first version of Composio failed — not because the market didn't exist, but because friendly customers gave him false hope. His fix? Talk only to strangers who would tell him his product sucked. The result: 100,000 developers and a $30M Series A in under a year.

Key Facts

• Total developer users: 100,000+ (Soham Ganatra)
• Series A raised: $30M from Lightspeed (Soham Ganatra)
• Time to raise Series A: ~2–3 weeks (Soham Ganatra)
• Customer calls per day at peak: 10–20 calls/day for several months (Soham Ganatra)
• Seed capital raised before first pivot: ~$4–4.5M (Soham Ganatra)

Why Friendly Customers Almost Killed Composio

Friendly customers signal false product-market fit. They accommodate founders rather than reject them honestly, causing teams to spend months building for one polite enterprise instead of finding real demand. Soham learned this after six months building for Glean — only to realize the core tech wasn't ready.

Composio's origin story is a cautionary tale about warm intros. After raising ~$4.5M in seed funding and leaving his role at Bureau, Soham spent six months building an AI-powered auto-integration product. He landed Glean as an early customer through his network — a real enterprise willing to pay six figures and more. It still failed.

The problem wasn't the market. It was the signal. Friendly customers don't tell you your product is broken. They suggest tweaks, ask for small changes, and stay polite. Soham describes it as people 'trying to do their best to accommodate you.' But accommodation is the enemy of honest feedback.

When Soham pivoted to Composio's current form — integrations specifically designed for AI agents — he made one deliberate rule: no more friendly customers. Every new user had to be a stranger who would tell him the truth.

"A friendly customer won't tell you this doesn't make sense. I'm not going to use anything even close to this. They'll be like, okay yeah, maybe if you do this, maybe we can use it. But you end up spending six months." — Soham Ganatra

"If you want real PMF you don't want people to use you because they know you. They should be just so sad about not getting anything out there that as soon as they see you, they use you." — Soham Ganatra

The Pivot: Building Integrations for AI Agents

Composio's pivot came from a problem Soham discovered while building the failed product: connecting AI agents to real-world tools like Gmail or Salesforce was deeply painful. Authentication, JSON schema design, and reliability all had to be rebuilt from scratch for how agents — not humans — consume software.

While building coding agents in mid-2023, the Composio team ran into a new category of problem. Tool calling — the mechanism that lets LLMs interact with external systems — was unreliable, poorly documented, and painful to implement. Nobody else was talking about it yet because almost nobody was building with AI agents at that stage.

Soham explains the difference between legacy integrations and agentic ones: 'An agentic Gmail integration would be designed entirely differently than a normal Gmail integration designed for humans. The way an agent navigates Gmail is pretty different.' For example, a long email thread that a human can scroll through will cause an LLM to break — it simply can't process that much context.

The core insight was that integrations for AI agents need to be purpose-built: optimized JSON schemas, reliable authentication flows, and tool designs that account for how LLMs actually consume data. Composio became the infrastructure layer that solves all of this so developers don't have to.

"Tool calling was not super stable. Apart from a lot of other problems we faced while building those agents, this was one of the interesting problems we found and I was like, okay, let's just go solve this problem right now." — Soham Ganatra

• Tool calling was unstable and poorly documented when Composio launched.
• Authentication (e.g., OAuth for Gmail) alone could take days to implement correctly.
• Agentic integrations require different API design than human-facing integrations.
• LLMs break on long-context responses that humans handle easily.

How Composio Got Its First 1,000 Developers

Composio's first users came from a viral Reddit post about improving tool calling accuracy, plus targeted Discord outreach in AI developer communities. Soham posted technical content that solved real problems — not marketing copy — and followed up with 10–20 cold customer calls per day to turn traffic into product insights.

With only ten integrations live, Composio needed users who would engage honestly. Soham's approach was to go where developers were already struggling. He published a detailed technical blog on improving tool calling accuracy — a topic every agent developer was wrestling with — and seeded it into Reddit and Discord communities. It went viral.

The CrewAI founder shared it on Twitter, calling it one of the coolest things he'd seen that day. That single mention drove a wave of signups. Meanwhile, Soham was in the trenches: 'I ended up booking ten to twenty calls a day for a couple of months.' Not all showed up, but enough did to generate continuous product signal.

He also ran programmatic SEO — creating landing pages for integration pairs like 'LangChain with HubSpot using Composio' — to detect real demand. When users arrived and found the integration didn't exist yet, he used that as a signal to prioritize the roadmap. He eventually took those pages down, but they served their purpose early on.

"We basically wrote a really nice blog and some code structure and some examples out there on GitHub. We put it out and said, hey, here's some good ways you could improve your code. It just hit the right audience. It went viral on Reddit." — Soham Ganatra

"I started just basically talking to a lot of them. I averaged more than ten calls a day for a very long time." — Soham Ganatra

Scaling from 1,000 to 100,000: MCP and Hype Cycles

Composio grew slowly until September 2024, when LLM quality improved enough for real production agents. A second wave hit in early 2025 when MCP (Model Context Protocol) created a new developer hype cycle. Soham's advice: identify the right hype cycles in AI and build compatibility before the wave peaks.

Until mid-2024, Composio had website traffic but few active users. The reason was simple: LLMs weren't good enough to power real-world agents yet. 'This number started increasing sometime in the second half of last year. When models just became better and you could actually build real world agents that kind of worked,' Soham explains.

The second inflection came with MCP — Model Context Protocol — which Anthropic introduced as a standard for connecting AI clients to external servers. MCP made software extensible in a new way: users could connect any MCP client (like Cursor) to any MCP server (like Notion or Jira) based on their own preferences, not just what the app natively supported. Composio added MCP compatibility, and the hype cycle brought a fresh wave of developers.

Soham's broader point is strategic: 'You have to latch on to the good hype cycles in AI. If the hype curve is right and people actually find value from what you're building in that specific hype cycle, it's a really cheap way to get a lot more users.'

"You have to latch on to the good hype cycles in AI. Because they're one of the good ways you can find people. If you're actually latching on to the right hype cycle and people actually find value from what you're building, it's a really cheap way to get a lot more users." — Soham Ganatra

Raising $30M from Lightspeed in Three Weeks

Composio closed a $30M Series A with Lightspeed in roughly three weeks by combining strong customer advocacy, a defensible infrastructure narrative, and a pre-existing relationship with the partner. Soham practiced his pitch using ChatGPT to analyze transcripts and identify weak answers before each round of meetings.

By early 2025, Composio had rapid user growth and needed capital to hire AI researchers. Soham timed the raise for January-February — when VCs are most active at the start of a new year. The process moved fast because the fundamentals were strong: customers were vocal fans, and the product sat at a critical infrastructure layer.

The pitch centered on a big market thesis: as foundation models move up the stack and natural language replaces app UIs, Composio becomes the de facto layer connecting users to any software without them needing to open those apps directly. 'At some point we can break into the monopoly of the application layer,' Soham explained.

His preparation tactic was unconventional: 'I ended up practicing with ChatGPT. I had all my transcripts with all the VCs in a single ChatGPT session and I used to say, what did I say wrong? Where should I have improved? What are the questions I was not prepared for?' The Lightspeed relationship was warm — his co-founder knew the partner personally — which compressed diligence time significantly.

"I ended up practicing with ChatGPT. I had all my transcripts with all the VCs in a single ChatGPT session and I used to say, what did I say wrong? Where should I have improved? What are the questions I was not prepared for?" — Soham Ganatra

"A lot of our people we were trying to raise from ended up reaching out to a bunch of customers and they just heard some insane, amazing things. They were like, okay, this is the craziest customer calls we've been to." — Soham Ganatra

• Time the raise when VCs are most active — early in the new year.
• Customer references are your strongest signal — make them vocal advocates.
• Use ChatGPT to debrief after every VC meeting and sharpen weak answers.
• Warm relationships with partners dramatically compress deal timelines.
• Valuation is subjective — story quality determines the range.

Soham's Contrarian Take: Don't Ask Users What They Want

In AI, asking users what they want produces 'faster horses.' Soham argues that if you have strong product intuition, you should build first and show users — because they can't articulate demand for things that don't yet exist. The iPhone didn't come from customer research.

Soham's most contrarian insight challenges the sacred startup principle of customer discovery. In AI especially, he argues, users lack the mental model to evaluate what doesn't exist yet. 'They would have just called bullshit on it. This seems too good to be true.' They might say yes to please you, or no because they can't imagine it working — neither answer is useful.

This doesn't mean ignore customers. It means sequence it differently: build a working prototype, demonstrate it, and watch the reaction. The 'eyes lighting up' moment is your real validation signal — not a survey response or a polite conversation.

Soham also admits he still doesn't feel confident Composio has true product-market fit. 'With AI, it's very difficult. The market is changing at such a rapid rate. It's very hard to say I actually have PMF for a very long time.' In a fast-moving space, PMF may be a moving target — something you chase in pockets rather than declare once and hold.

"My hunch with AI space is that you can't validate problems by asking people, because they won't get it. If you just go build it and show it, they would really get it." — Soham Ganatra

"If you're building a new product and you have very strong intuition on why it will work, I would say this is one of those times where multiple products like the iPhone could come out." — Soham Ganatra

Legacy Integrations vs. Agentic Integrations