Town's approach to AI at work: finding opportunities and doing the task

Greze said Town aims to work with people to find more opportunities to use AI in their jobs, starting with meetings and daily preparation.

He said the product cannot be just a text box; it has to show value directly and do the work for the user. Running it costs about a dollar a month, but the one month a year it finds something can save hundreds or thousands of dollars.

Greze said humans are slow routers and that the future will need far less human involvement in the loop.

Greze said incumbents could reach hundreds of millions in revenue before a new entrant because tax is sold only once a year, which makes it impossible to be a daily-active-use product.

He added that people rarely talk about their tax prep, and the only time they do is after filing, when they are not shopping for a new preparer. A friend in edtech faces the same constraint, since schools buy software only in two windows a year.

Greze said the team forgot that limit when it chose tax.

Greze said the team could not pivot after telling customers and signing partners, so the decision had to be made in a window of roughly a month around October 15th.

He contrasted this with continuous businesses, where founders can keep testing ideas for weeks. The seasonal deadline forced a faster decision, because a year later they would not have the capital to change direction.

Greze said most businesses run on email and calendar, especially external roles such as sales, business development and recruiting. The team concluded the market was large and global.

The host compared enterprise AI adoption to the iPhone, noting that Steve Ballmer dismissed the $800 phone and BlackBerry soon collapsed. Greze agreed with the framing and added that people had been exposed to AI in personal settings but not at work.

Greze said enterprise sales run into long diligence cycles and security requirements, while SMB and mid-market buyers have fewer of them, so adoption is much faster.

The current plan is self-serve growth in SMB and mid-market, with enterprise security, compliance and finance features added later.

Greze described the product as a token allocation machine that offers humans options for where to spend tokens. As tokens get cheaper, uses that were never worth human time become viable, such as per-meeting prep notes.

He gave the example of a restaurant owner whose wholesalers send price lists once a month or quarter. An AI checking those lists costs about a dollar a month, and the one month that finds a change could save hundreds or thousands of dollars.

The host said that asking a restaurant owner about an executive assistant gets a blank look, while asking about saving money on lettuce gets an immediate yes.

The point ties back to the earlier question of building for the user first and selling to enterprises later.

Greze said Town does not look at user data. Instead it reviews aggregate metrics to find power users, interviews about ten of them in a segment such as restaurants, and turns the common patterns into prototype automations called routines.

New users get templates based on their inbox, and as AI improves the library has grown from about five routines to roughly fifteen for restaurant owners. Greze said the process is moving from human speed to AI speed.

Greze used the example of finding someone at a prospect company. In Slack or Teams, every employee would have to read and answer, but Town's agents can answer immediately after the company decides to trust AI to mediate privacy for such requests.

He argued the economic benefit outweighs the risk and that companies should be comfortable with a future with far fewer humans in the loop.

He said the bottleneck is people who can tinker a solution to their own problem, and that a product cannot just be a text box.

The speakers said people who tried AI a couple of years ago judge it by that version, even though the product keeps improving. Greze gave the example of a Harvard professor who had an AI rewrite her dissertation with current content and was amazed.

Silicon Valley judges where the technology could go, while many others judge it as it is today.

The speakers said agent traffic has vastly surpassed human traffic. They described the old covenant in which Yahoo gave free stock quotes in return for ad impressions, and said agents have broken it.

The speakers said people with more time than money use coupon and price-tracking tools, while people with more money than time do not. They argued agents could give the time-poor the same tools.

They added that coupon and breakage models would change as benefits are used fully.