Leaders in Agent-Led Growth
TL;DR: MeetStream is the meeting bot API for real-time AI agents: one API to join, record and stream Google Meet, Teams and Zoom calls. Its customers are developers, and more of them each month ask a coding agent which API to use. MeetStream was losing that decision four times in five. Lightsage showed it why. After the changes, agents recommended MeetStream four times more, and the integrations they built worked nine times in ten on 36% fewer tokens.
"We had more context about our own product than any tool could give us. What we didn't have was a testbed: something that ranks us against the alternatives on every prompt, every day, so we can see what moved. Lightsage is that."
Sidhdharth, MeetStream Co-Founder and CEO
The problem: the buyer is an agent, and it was choosing someone else
MeetStream's customer has always been a developer: someone building a sales assistant, a note-taker or a voice agent, who needs a bot in the call and doesn't want to write three platform integrations to get one.
In 2026 that developer increasingly opens a coding agent and types "add a bot that joins our Google Meet calls and answers questions." The agent picks an API, reads its docs and writes the integration. The developer may never open MeetStream's site.
MeetStream saw this early. By the autumn of 2026 it had published everything an agent needs to find and use an API: an MCP server, a CLI, an agent skill, llms.txt and an OpenAPI spec. And still the agents chose someone else.
Lightsage asked a coding agent the question a developer would ask: which meeting bot API should I use for a bot that speaks and answers questions during calls? The agent read MeetStream's homepage and picked the established name anyway, calling it "the safer default." One win in five. Asked to build a calendar integration on MeetStream, the same agent produced a working script four times in ten.
The product wasn't the problem. The missing piece was a view of the decision: what the agent read, what it weighed, where it fell off.
What Lightsage showed
Lightsage put the same questions to coding agents over and over, and showed MeetStream how each one reached its answer: the pages it read, the reasons it gave, the point where a build went wrong.
The pattern was consistent. The capabilities that made MeetStream the better answer for a voice bot were in the product, not spelled out on the page the agent read. On the build task, most agents wrote their own scheduling code for a job MeetStream's API already does; the docs they read didn't point them to it.
That turned a general worry into a short list. MeetStream made a small set of changes to how its site and docs present the product to an agent. Lightsage ran the same tests again.
Results
4x more recommendations. On the same question, MeetStream went from one win in five to four. Every winning answer cited the capabilities MeetStream had made visible, in the page's own words. All five runs were confirmed to have read the updated page.
Working integrations, 40% to 90%. Nine in ten of the agents that read the updated docs built the calendar integration on MeetStream's native endpoint instead of working around it. It took them no extra time.
Shorter, cheaper runs. On a quickstart task, agents reached a working voice agent on 36% fewer tokens: 693k per run down to 444k. Tool calls fell from 20 to 17. Every run passed.
Key takeaways
- Being usable isn't the same as being chosen. MeetStream had published everything an agent needs to find and use its API, and agents still picked the established name four times in five. The recommendation is a separate decision, and it has to be measured on its own.
- Agents decide from the few pages they read. The capabilities that made MeetStream the better answer existed. They weren't on the page the agent opened. Saying plainly, on that page, what you do better changed who the agent recommended.
- Agents quote you. Every winning answer cited the updated page in its own words. What you write for agents is what they say about you.
- If agents can't find a feature, they build around it. Most agents wrote their own scheduling loop for a job MeetStream's API already does. One pointer in the reference turned four working integrations in ten into nine.
- Re-run the same test. Two small changes, the same prompts before and after. That is how MeetStream knows which change did what, and how the next one will be judged.
What's next
MeetStream now tracks these prompts every day across Claude Code, Codex, ChatGPT, Perplexity and Google AI Overviews. It is applying the same approach to its CLI, MCP server and agent skill, so an agent that finds MeetStream can also build with it on the first try. This is what Agent-Led Growth looks like in practice: measure the agent, find the why, ship the fix.
MeetStream is the meeting bot API built for real-time AI agents. Lightsage is the Agent-Led Growth platform. It shows API companies how agents discover, evaluate and use their products, and what to fix.
