Sonnet
Summer 2022InactiveMeeting notes and CRM, automated
Pressure-test this opportunity
Explore the risks and possibilities with a prompt for ChatGPT, Claude, or your agent.
Meeting notes and CRM, automated
Explore the risks and possibilities with a prompt for ChatGPT, Claude, or your agent.
Sonnet was a San Francisco meeting assistant founded in 2022 by UC Berkeley roommates Ted Chai and Dylan Feng. The S22 company combined bot-free recording, transcription, configurable AI notes, pre-meeting research, follow-up drafting, and an automatically updated relationship database.[1]
The public record points to a product-market-fit problem, not a lack of technical ambition. Sonnet sold a broad workflow while summaries were becoming bundled meeting-platform features. Its CRM also asked customers to adopt another system of record. Chai now describes the experience as a “scramble for product market fit.”[2] Both founders report a 2024 exit, but do not identify the acquirer or terms.[7][8]
Chai and Feng met as roommates at UC Berkeley. YC says Chai left college early after work at Uber and the U.S. Department of State, while Feng graduated in two years and conducted research at the Berkeley Artificial Intelligence Research lab.[1] That pairing gave Sonnet a plausible founding advantage: one founder had operating and policy exposure, while the other had recent machine-learning research experience.
The product came from a specific frustration. In Sonnet's Product Hunt launch, Chai wrote: “I found that I was often dropping the ball on important information and tasks mentioned on calls.” He said he would replay entire recordings before follow-up meetings, then described the intended product as one place where a team’s conversations became “concise, actionable, and searchable.”[3]
Feng’s public account emphasized craft more than a new technical primitive. He wrote that the team’s favorite answer to competitive questions was “the amount of care we put into the engineering and design of Sonnet,” adding that six months went into testing details and responding to complaints.[4] Sonnet’s differentiation depended on better execution across a bundle competitors could reproduce.
Chai says Sonnet was acquired in 2024; Feng dates his exit to October.[7][8] No public source located the buyer, price, integration, or customer outcome. Chai later joined Recall.ai, but that sequence does not establish it as the acquirer.[2]
Sonnet treated a meeting as a workflow rather than a transcript. Before a call, it assembled information about participants and sent reminders. During the call, its Mac application captured audio from the device without adding a visible bot to Zoom or another meeting. Afterward, it produced a transcript, generated notes in a user-selected structure, identified follow-up material, and added interaction history to Sonnet’s own CRM.[1]
Generic summaries often missed fields that mattered to a salesperson, recruiter, or researcher. Sonnet let users instruct the model what to capture. Bot-free recording removed the awkward extra participant in every call.
The built-in CRM was the boldest choice. Chai said Sonnet deliberately did not merely export data to an existing CRM because an internal system could be “truly automated and organized, instead of being a data dump through an integration.”[3] This made the end state coherent, but enlarged the adoption burden. A customer had to trust Sonnet to record sensitive conversations, generate accurate notes, and become a durable relationship database.
The launch exposed narrower constraints too. Chai confirmed the product was Mac-only and had no web application. One Product Hunt commenter reported two automatically scheduled meetings were not recorded. That is one public complaint, not evidence of a general reliability failure, but missed capture is a severe failure mode for a product whose promise is perfect organizational memory.[3]
Sonnet’s product language pointed toward meeting-heavy professionals and small B2B teams, particularly sales, recruiting, and customer-facing operators. Its free tier reduced trial friction, while tiered pricing implied an eventual SaaS model.[3]
No credible contemporaneous market-size estimate or revenue disclosure was located. Conflicting third-party directories are not reliable enough to establish traction. Existing notetakers and native platform summaries provide the safer demand signal.
Sonnet competed on three axes: capture without a bot, notes that followed custom instructions, and automatic conversion of conversations into relationship records. Specialist competitors could copy the first two. Platform owners had a stronger distribution advantage. Zoom now generates structured summaries from speech-to-text data inside paid Workplace accounts and distributes them through email and chat.[5] Its templates cover one-on-ones, standups, customer success, user feedback, and other meeting types.[6]
Sonnet’s CRM could have accumulated proprietary context, but also competed with established systems. The product sat between meeting platforms that owned capture and CRMs that owned customer workflows.
Sonnet offered a free tier and tiered paid pricing, but the accessible launch material does not preserve prices, conversion, revenue, retention, or gross margin.[3] YC participation establishes institutional backing, but public directories disagree about total capital and should not be treated as audited funding data.
The product implies costs for transcription, model inference, storage, and desktop development, but Sonnet disclosed no economics. Bundled summaries pushed the visible price of the core feature toward zero, leaving better notes, cross-platform memory, or CRM automation to carry monetization.
The launch produced 389 Product Hunt upvotes and a number-three daily ranking.[3] Keywords AI founder Andy Li wrote that Sonnet recovered customer context and compared it favorably with Fireflies and Otter. This is testimony, not verified company traction. No cohort, active-user, paid-seat, or revenue figures were found.
Sonnet saw that meeting value begins before a call and continues after it, then built every adjacent step. That made product-market fit harder to isolate. Notes buyers had to assess a new CRM; CRM buyers had to accept a young recorder.
The attempted remedy was integration through ownership: Sonnet would control capture, notes, follow-up, and relationship history. The outcome was an elegant loop with a large replacement surface. Chai’s later description of a product-market-fit scramble is the clearest first-party assessment available.[2]
The structural problem was distribution. A specialist had to persuade each user to install recording software, grant access, and change behavior. Zoom could place summaries inside the meeting product and include them in eligible paid plans. It now turns transcripts into summaries, action items, editable records, and role-specific templates without a separate purchasing motion.[5][6]
Sonnet tried to move above that commodity layer with custom notes and relationship memory. The difficulty was that its CRM competed with systems customers had already standardized on. The product therefore faced pressure from below and above at once.
Bot-free desktop capture improved meeting etiquette, but local capture added platform coverage and reliability work. The launch’s reported missed recordings show the severity of the edge case even if they do not establish its frequency.[3] A note-taking tool can survive an imperfect summary that users edit. It cannot easily recover a conversation it failed to record.
The counterargument is that the category supported large independent companies. Sonnet may have needed more time, capital, or a narrower buyer. The record cannot distinguish those possibilities. Its differentiation was spread across craft, workflow breadth, and an internal CRM rather than concentrated in one distribution advantage.