Aiolum Browse stories
All stories
AI agents for purchasing, inventory and sales work at wholesale distributors · United States

Lark: Product, team and how it was built

Small teamFundedB2BWholesale distribution

Lark (AI agents for purchasing, inventory and sales work at wholesale distributors, United States): Live with a published customer story, November 2026. Evidence grade E: no revenue figure was checked. Team: 2 co-founders; Y Combinator lists a team size of 2 in November 2026. Launched September 2026; still running.Added 6 November 2026

Outcome
Live with a published customer story, November 2026
Evidence
E no revenue figure was checked
Launch
September 2026
Status
Still running
Team
2 co-founders; Y Combinator lists a team size of 2 in November 2026
Outside funding
Y Combinator, Fall 2026 batch; amount not published
Main channel
Not stated
No revenue figure

What is not verified

No revenue figure for Lark was found. We looked for a payment-processor profile, a founder interview with figures, a Y Combinator launch post and a filing. The customer story gives a customer outcome (about 60 hours a month saved), which is not Lark revenue. Jason Wang's Y Combinator profile says that his previous company, Datost, also backed by Y Combinator, "grew ARR to 400k in 8 weeks"; that is his own statement about a different company and was not checked. A reader could look for a launch post, a funding announcement or an interview with the founders. This story is grade E: the business is real and its product, price and customers are described from public pages, but no revenue figure was checked. Aiolum shows it as an idea worth studying, not as proof that it makes money.

Lark sells AI agents, programs that use a large language model to read data, decide on a next step and take an action, to wholesale distributors. The agents connect to the customer's ERP (enterprise resource planning system, the database where a distributor keeps items, stock, orders and prices), such as Epicor Prophet 21, Infor CSD or NetSuite. They draft purchase orders (POs, the orders a distributor sends to its vendors), forecast demand, flag stock problems, enter customer orders and draft quote replies.

Jason Wang and Michael Wang founded the company in San Francisco in 2026 as LarkAI Technologies, Inc., and Y Combinator lists it in the Fall 2026 batch with a team of 2. The site's terms are dated 16 September 2026, and the Internet Archive first saved the site on 29 September 2026. The company publishes one customer story: Interline Hardware, a hardware wholesaler with more than 60,000 items.

No revenue figure was found, so this story has grade E. The site publishes no prices; customers start with a pilot after a call.

Lark: Live with a published customer story, November 2026. AI agents for purchasing, inventory and sales work at wholesale distributors, United States, team of 2 to 10, evidence grade E.
The story in one card. Share it with the address of this page.
01

How much revenue does Lark make?

Each row shows the evidence grade and the source numbers from the list at the end of this page. The grades are explained in How Aiolum verifies.

A two-founder Y Combinator company with six agents, one named customer and no published revenue or prices
FactValueGradeHow we knowSources
Founded2026, San FranciscoCY Combinator company page, read 6 November 2026.[3]
Team2 people: Jason Wang (CEO) and Michael Wang (co-founder)CY Combinator company page lists team size 2 and both founders.[3]
FundingY Combinator, Fall 2026 batchCY Combinator company page; the home page says "Backed by Y Combinator". No amount is published.[1, 3]
Earliest public date of the siteterms last updated 16 September 2026; first archive capture 29 September 2026CTerms of service page and the Internet Archive capture index.[5, 6]
Agents offeredPurchasing, Demand forecasting, Inventory monitoring, Order entry, Sales operations, Data analysisCHome page and FAQ. The FAQ adds that the company can also build customized agents.[1, 4]
ERPs namedEpicor Prophet 21 and Eclipse, Infor CSD, Oracle NetSuite, Acumatica, SAP Business One, Microsoft Dynamics 365 Business Central, Sage 100/300, Spire, DDI System InformCFAQ, which says this is not a complete list. It also names Outlook, Gmail, Google Drive, SharePoint, OneDrive, Dropbox, QuickBooks and HubSpot.[4]
Control over ERP changesonly through approved workflows with limitsCFAQ example: "POs under $5,000 for approved vendors go through on their own; bigger ones wait for a buyer." Every read and action is logged.[4]
Named customerInterline Hardware, more than 60,000 inventory itemsCCompany's customer story. The customer's head of purchasing is quoted by first name only.[2]
Customer's time savedabout 60 hours a monthCCustomer quote in the company's story: the reorder list took three hours a day and now takes about 30 minutes. A customer outcome, not Lark revenue.[2]
Setup timeone to two months for most setups; under one month for InterlineCFAQ and customer story.[2, 4]
HostingGoogle Cloud in the USCFAQ. Customer data is not used to train models, the FAQ says.[4]
Time from launch to the customer storyabout 2 monthscalcCalculated: September 2026 to November 2026.[2, 5]

How to read this table: a grade C value is the company's own statement on its site or its Y Combinator page. A "calc" value was computed on this page. No value is grade A or B, and no revenue figure exists.

Lark prices read on 6 November 2026 from the company's site (grade C): no list price is published
PlanPer month, billed yearly (US$)Per month, billed monthly (US$)Limit
Pilot after a booked callno list pricesetup of one to two months; more agents can be turned on later

Every page offers "Book a meeting" for a 30-minute call. The FAQ says most setups take one to two months and start with a pilot in which Lark and the customer's team map data, items, vendors and rules. No plan names, prices or limits appear on any page that was read, and the site has no pricing page.

02

When did Lark launch, and how did it grow?

From incorporation to a published customer story in about two months. Each entry carries its grade.

  1. 2026CJason Wang and Michael Wang found LarkAI Technologies, Inc., a Delaware corporation, in San Francisco. [3, 5]
  2. 16 September 2026CThe terms of service are last updated; the privacy policy takes effect on 17 September. [5]
  3. 29 September 2026CThe Internet Archive saves larkagents.com for the first time. [6]
  4. Fall 2026CLark is in the Y Combinator Fall 2026 batch. [3]
  5. November 2026CThe site offers six agents and publishes a customer story about Interline Hardware. [1, 2]
03

Why did Lark work?

Four causes, in the order they mattered. Each one is a reusable lesson.

01

One industry, one set of systems

Lark sells only to distributors, and its demos use their words: backorders, reorder points, vendor lead time, price lists per customer. The FAQ names more than ten ERPs that distributors use. A narrow buyer lets a two-person team write agents that know the data layout of those systems.

02

On top of the ERP, not instead of it

Replacing an ERP is a project of years for a distributor. Lark reads and writes the existing ERP, so the buyer's only change is a pilot of one to two months.

03

Approval limits as the safety rule

The customer approves each workflow once with limits, for example automatic POs below US$5,000 for approved vendors, and Lark logs every action. This answers the main fear about agents that change business records.

How it was built

These details come from the company's own pages. They are grade C. [4, 5]

  • Hosted on Google Cloud in the US; data encrypted in transit and at rest; credentials kept in a separate vault per customer that does not reach the AI model.
  • The terms say AI features use "third-party AI models"; no model provider is named.
  • The web app uses PostHog for analytics and session replay, according to the privacy policy.
  • Setup is done by the Lark team with the customer during a pilot.

Tools named in the sources

Each one opens its page in the Tools tab, with every story that uses it.

04

Who tried this before Lark, and why did they fail?

Three companies in our database of failed startups tried to digitize distribution or small-business inventory work and failed. The values below are from that database and are not independently verified.

Died 2015 to 2022Burned US$60M

ShopX

The promise

Digitize India's corner shops and cut out distributor middlemen.

Why it died

The distributors who controlled supply did not give way, and customer acquisition cost more than customers brought in.

Unit Economics
Died 2020 to 2024Burned US$10M

Kyte (B2B SaaS)

The promise

One platform for inventory, billing and CRM for small businesses in India.

Why it died

Low revenue per customer, high acquisition cost and long sales cycles.

Unit Economics
Died 2015 to 2022Burned US$340M

Zilingo

The promise

Supply-chain software, financing and inventory tools for fashion merchants and factories.

Why it died

It spent far more than it earned, and an accounting scandal ended investor trust.

Ran Out of Cash

The failed startups come from our database of 1,749 dead startups, retold here in our own words. Its entries are AI-assisted summaries of public sources, so these cards carry no evidence grade.

What changed

The predecessors tried to replace how distribution works, or to sell a new all-in-one system to many small buyers. Lark sells to distributors that already have an ERP and adds agents on top, so it does not fight the distributors or ask them to migrate.

Language models can now read customer emails, purchase orders in PDF files and messy ERP records, which made order entry and quote replies automatable without custom rules per customer. Whether this sells at a price that covers the one-to-two-month setup is not yet shown by any public figure.

05

Can one person copy Lark now?

Yes, for someone who knows one industry's systems.

The method is a narrow buyer, agents on top of the systems that buyer already uses, and approval limits. A solo founder can copy it, but setup by hand with each customer limits how many customers one person can serve.

Where it is still open: Other trade industries on old ERPs: food service distribution, building materials, auto parts, medical supply.

  1. 1Pick an industry that runs on a few old ERPs, and learn their data.
  2. 2Automate one daily task that the buyer can measure in hours, such as the reorder list.
  3. 3Let the customer set limits for automatic actions, and log every action.
  4. 4Publish a named customer story with the time saved.

What could go wrong for Lark?

The risks are the mirror image of the causes above.

What is not verified

No revenue figure for Lark was found. We looked for a payment-processor profile, a founder interview with figures, a Y Combinator launch post and a filing. The customer story gives a customer outcome (about 60 hours a month saved), which is not Lark revenue. Jason Wang's Y Combinator profile says that his previous company, Datost, also backed by Y Combinator, "grew ARR to 400k in 8 weeks"; that is his own statement about a different company and was not checked. A reader could look for a launch post, a funding announcement or an interview with the founders.

Trust signals

The only testimonial is quoted as "Mike, Head of Purchasing, Interline Hardware", with a first name only. The customer company is named. The Y Combinator backing is confirmed by the Y Combinator company page.

Setup by the company

The site says "We handle the setup with your team", and the FAQ says "We can also build customized agents for your workflows". The product is software, but each customer needs a pilot of one to two months set up by the Lark team.

Website tracking

The privacy policy says the website uses "third-party visitor identification that may associate activity with a company" and Google Ads measurement, and that the app's session replays can include customer business records.

Competition and model providers

LineSight, another story on this site, sells AI order entry to metal service centers, and FastQuery an AI assistant to hardware stores. Lark's terms say it uses "third-party AI models" without naming them, so its dependence on one provider cannot be judged.

Sources

The grade next to each source is the best grade that the source can support. "Read in full" means that the text of the page was downloaded and the numbers were checked in it.

  1. C Lark home page: Specialized AI Agents for Distributors
    Read in full on 6 November 2026. Source for the agents, the approval limits and the Y Combinator backing.
  2. C Lark customer story: How Interline Hardware manages inventory and purchasing with Lark
    Read in full. Source for the named customer, the item count, the setup time and the time saved.
  3. C Y Combinator company page: Lark, AI agents that automate workflows for distributors
    Read the company data embedded in the page: year founded, team size, batch, location and founder profiles.
  4. C Lark FAQ
    Read in full. Source for setup time, the ERPs named, the control rules and the hosting.
  5. C Lark Terms of Service, last updated 16 September 2026, and Privacy Policy, effective 17 September 2026
    Read both in full (the privacy policy is at https://larkagents.com/privacy). Source for the legal entity, the third-party AI models, PostHog and the tracking.
  6. C Internet Archive capture index for larkagents.com
    Read the capture list. The first capture is dated 29 September 2026.
  7. C Database of failed startups, local export
    Local export. AI-written summaries, not independently verified. Source for the failed predecessors.

Run this company? Send corrections or a verified figure to aiolum@livyden.com

Back to all stories