# General Employee

AI-powered assistants for SMBs

- Team: Paul Sheridan (founder)
- Invested: 2022
- Links: [Website](https://lynq.ai), [LinkedIn](https://ie.linkedin.com/in/paulfsheridan), [X](https://x.com/paulfsheridan), [GitHub](https://github.com/paulfsheridan)
- Field: Making and talking

## The problem: How do you answer every call when you're the one under the sink?

### How the front desk works

The U.S. has 36.2 million small businesses, and they account for almost half of private sector employment. Most business establishments in the country have no paid employees at all, and most of those are self-employed people running sole proprietorships. In a business like that, whoever does the work also answers the phone, keeps the calendar and chases the invoices, usually all at once.

An AI phone assistant usually works as a chain. Speech-to-text turns the caller's words into text, an agent decides what to do, and text-to-speech speaks the reply. The other design uses one model that hears audio, decides what to do and answers in speech, all in a single session. Either way, the assistant is only useful if it can reach the business's own tools, like its calendar.

### Why it is hard

**A fifth of a second.** People are very fast at taking turns. Across ten languages, from indigenous communities to major world languages, speakers avoid talking over each other and keep the silence between turns short. In every language studied, the most common gap before a reply fell between 0 and 200 milliseconds. A phone assistant that pauses to think sounds broken, and a chain of three models has to fit its work inside that gap.

**Hearing it right.** Speech recognition isn't equally good for everyone. In one test of five commercial systems, the average word error rate was 0.35 for black speakers and 0.19 for white speakers. On the phone, a misheard street name or appointment time means a technician driving to the wrong place.

**Confidently wrong.** Language models sometimes make things up, and a business's assistant speaks for the business. Air Canada's support chatbot invented a bereavement fare policy, and a tribunal ordered the airline to pay damages and honour it. An assistant that quotes prices or promises appointments has to know where its knowledge ends.

**Rules on the line.** Following up is regulated too. U.S. law limits calls to customers that use automatic dialers, artificial or prerecorded voices and text messages. An assistant that texts confirmations and chases leads has to play by those rules as well as be polite.

### What General Employee is after

General Employee builds an AI receptionist for local businesses such as auto shops, med spas, dentists, plumbers and HVAC firms. It answers the phone, chases leads and books the work around the clock.

It doesn't replace the owner's line. The business keeps its number and sends over the calls it would otherwise miss: after hours, at weekends, or whenever the line is busy.

### How they go at it

**Book, don't just answer.** The assistant connects to the calendar the business already uses, offers real openings, books the job during the call and sends the customer a confirmation. It asks screening questions that follow the business's own booking process.

**One conversation, every channel.** The same assistant answers the phone, replies to texts and chats on the business's website. It notices which language a customer speaks and answers in that language.

**Knowing when to hand off.** When it doesn't know an answer, it takes a message, tells the caller someone will follow up and flags the conversation for the owner. Teach it the answer once and it handles the question next time. It doesn't pretend to be a person, and higher plans add warm transfers to a human with an AI briefing.

**Guardrails on the outbound.** Texts it sends can respect registration, opt-outs and quiet hours. Calls, texts and customer details are never used to train AI models.

### Still open

How fast is fast enough? Languages differ in their average gap between turns, though only within about 250 milliseconds of the cross-language mean. Newer full-duplex voice models can listen and speak at the same time, which lets a caller keep talking while the system works. Whether that feels natural on a crackly phone line is still being worked out.

Can speech recognition work equally well for every caller? The racial gap in that study held even when black and white speakers said identical phrases, which points to the acoustic models themselves. Suggested fixes include training on more diverse speech, including African American Vernacular English.

### Words used here

- **sole proprietorships**: Businesses owned and run by one person, with no legal separation between owner and business.
- **Speech-to-text**: Software that transcribes spoken audio into written words.
- **text-to-speech**: Software that reads written text aloud in a synthetic voice.
- **word error rate**: The share of words a transcription gets wrong, counting substitutions, deletions and insertions.
- **warm transfers**: Handing a live call to a person along with a summary, so the caller doesn't have to start over.
- **full-duplex**: Able to listen and speak at the same time, the way people do on a call.

## About General Employee

General Employee builds AI-powered assistants that automate routine tasks for fund managers and analysts. The platform specializes in workflows and search tailored to the finance industry, with a focus on venture capital firms. The company also offers AI agents for SMBs designed to run operations continuously without human intervention.

## Sources

1. [New Advocacy Report Shows the Number of Small Businesses in the U.S. Exceeds 36 million](https://advocacy.sba.gov/2025/06/30/new-advocacy-report-shows-the-number-of-small-businesses-in-the-u-s-exceeds-36-million/), U.S. Small Business Administration, Office of Advocacy
2. [About Nonemployer Statistics](https://www.census.gov/programs-surveys/nonemployer-statistics/about.html), U.S. Census Bureau
3. [Voice agents](https://platform.openai.com/docs/guides/voice-agents), OpenAI
4. [Universals and cultural variation in turn-taking in conversation](https://pmc.ncbi.nlm.nih.gov/articles/PMC2705608/), PNAS (via PubMed Central)
5. [Racial disparities in automated speech recognition](https://pmc.ncbi.nlm.nih.gov/articles/PMC7149386/), PNAS (via PubMed Central)
6. [Hallucination (artificial intelligence)](https://en.wikipedia.org/wiki/Hallucination_(artificial_intelligence)), Wikipedia
7. [Telephone Consumer Protection Act of 1991](https://en.wikipedia.org/wiki/Telephone_Consumer_Protection_Act_of_1991), Wikipedia
8. [General Employee - Your AI Receptionist](https://generalemployee.com), General Employee