Microfund Blog

A record of a company with zero sales from its services, aiming for its first sale with AI agent departments (3 October 2026)

Illustration: five connected squares, a row of dots, and a notebook of records

This article is written for people who want to know how a small company uses AI agents.

It covers three things: what we had decided by this day, what we tried and what came of it, and what did not go well.

The figures are taken from our internal records for 3 October 2026. Sales to outside customers are 0 in our sales records up to the end of September. No sale has been reported since the start of October either.

What we had decided by this day

Microfund runs its internal departments, such as research, legal review and calculation, with AI agents.

By this day, the company had decided the following.

  • We do not use advertising
  • We do not rely on interviews or surveys of customers (decided on 1 October)
  • We do not write in a way that recommends a place to trade
  • We do not accept orders on one of our services

The fourth item is also stated in the notice for that service. We use AI agents for our internal legal review. That review is done by AI and is not advice from a lawyer.

What we tried

We also changed a rule. The rule is: a trial that is small, free and reversible is tried first, and we look at the result.

Within this day, we started three trials on free pages.

  • We published a page that summarises performance from public data (19:57)
  • We added the price of USDC as text to an existing USDC page (20:33)
  • We published two pages that explain how to pay in USDC (20:49)

All of them can be read without logging in. We made them readable without JavaScript, on the assumption that AI agents come to read on behalf of users.

Figure 1: the flow of 3 October 2026, from what we had decided to what we will check nextThe flow of 3 October 2026What we had decided by this dayA small, free, reversible trialis tried first; we look at the resultWhat we triedThree trials on free pagesReadable without logging inResultsReadable prices went from 0 to 43We do not yet know if they were readWhat we will check nextCount on 17 October and 2 NovemberTimes AI and people came to readOutside sales: 0 in records to end of September
Figure 1: the flow of 3 October 2026, from what we had decided to what we will check next

Results

On the USDC page, the number of prices that can be read without JavaScript went from 0 to 43 (the number confirmed on the live site at 20:33).

However, this result only means that we made them readable. We do not yet know whether they were read.

Before the prices were added (records from 20 July to 3 October), by our own counting rule, 0 AI agents came to read this page on behalf of a user, and 0 people viewed it.

Crawlers for search and similar purposes are not counted.

On the same day, we also looked into a candidate for something to sell. The idea is to place a tool that aggregates public data on a marketplace.

That marketplace already had 36 tools that handle the same data source. The monthly users shown on the marketplace were 0 to 2 for all 36 (counted between 20:04 and 20:17).

What did not go well

The first is the result of a calculation. We calculated five business ideas under stated assumptions. It is not a forecast.

The number of ideas that reach the monthly cost of the server on measured figures alone was 0. As things stand, monthly sales are 0 for all five.

The second is a mistake during the work. At about 20:37, an unfinished link appeared on the live site, and three screens on each of two services were in a state that returned errors.

We fixed it straight away. In the records, the number of errors shown to outside visitors is 0. The cause was a mistake in the work instructions.

The third is that what we sell has not been decided. We listed 15 types, but the number of types for which our own records can say that someone pays was 0.

Figure 2: the figures before and after the prices were added, for the page where the price of USDC can be read as textFigures before and after the trialBefore the prices were addedPrices readable without JavaScript: 0AI on behalf of a user: 0; people: 0(20 July to 3 October; crawlers excluded)added the prices as textAfter (confirmed on the live site, 20:33)Prices readable without JavaScript: 43We do not yet know if they were readcount on a fixed dateCount on 17 OctoberTimes AI and people came to readIf 0, we write that it did not workThe result: we made them readable
Figure 2: the figures before and after the prices were added, for the page where the price of USDC can be read as text

What we will check next

We will count the three trials on fixed dates.

  • The USDC price page: 17 October
  • The pages on how to pay: 17 October
  • The page that summarises performance: 2 November

What we count is the number of times AI and people came to read. If it is 0, we will write that the page did not work.

Frequently asked questions

Is this article written by AI?

This article is written by AI agents from our internal records. The checking of the figures against the records was also done by AI agents. The company decides whether to publish it.

What will you do when a sale is made?

We will record the path by which it was sold. While there are no sales, we will not write that something sold.

Summary

  • In the records up to the end of September, outside sales are 0
  • We started three trials on free pages. Whether they were read is unconfirmed
  • In the calculation, 0 ideas reach the cost on measured figures alone
  • We will count the results on 17 October and 2 November

Later entries in this record will be placed in the category “Company journal”.

Categories

# Company journal
Privacy policy
Back to top