Reading Room

How AI agent products get priced, whether per seat or per use, is a live question for anyone building or buying these tools, and this fintech newsletter flags one company testing it in public.

Perplexity, the AI search company, has launched a product called Teammates. The newsletter describes it as Perplexity's first always-on AI agent, meaning software that keeps working on its own without someone prompting each step. It is built, in the newsletter's words, as "a staff of isolated agents": separate AI workers that each handle their own task rather than one shared assistant juggling everything.

The newsletter says it plans to compare Teammates against competing agent products from three other companies: OpenAI, Meta, and xAI. Its headline argument is that per-seat pricing, charging a flat fee for every person who gets a login, is already running into trouble for a product shaped like this one. The logic the newsletter is gesturing at, though it does not spell it out in the free preview, is the mismatch between what a seat-based fee assumes (one person, logging in, using it) and what an always-on agent actually does (running continuously, with no single person's login driving the work).

That is as far as the free preview goes. The piece is part of a longer paid edition that also covers Visa's latest earnings in depth, and the newsletter keeps its actual reasoning on the Teammates pricing question, along with any numbers behind it, for paying subscribers. What is here is the question being raised, not yet the answer: can a flat per-person fee make sense for an agent that runs continuously rather than one person logging in to use it when they need it. The source text available does not go further than this, so this item stays short rather than guessing at the subscriber-only argument.

This is background only, useful if he ever considers paying for buy side representation during his own acquisition search, because it lays out exactly what that kind of broker does and what it costs.

A buy-side broker represents only the buyer, never the seller. Unlike a normal seller's broker, who works to get the owner the highest sale price, a buy-side broker runs the whole mechanical process of finding, checking, and landing a small or mid-sized business (SMB) that matches a buyer's exact criteria. The sequence runs in five steps: set the search criteria (industry, EBITDA range, that is profit before interest, tax, and write-downs, geography, and business model); run cold outreach to find deals that were never publicly listed, called the "hidden market"; screen the resulting leads' early financials; model the actual deal, meaning valuations, working capital, and structures like seller financing, earn-outs, and clawbacks; and coordinate the rest of the deal team, meaning accountants, transaction lawyers, and SBA (Small Business Administration) lenders.

Buy-side brokers charge in two main ways. The first is a monthly retainer plus a success fee: retainers run $2,500 to $7,500 or more a month depending on the size of the target business, and fund dedicated search hours, with a success fee as a percentage of the final purchase price paid at closing, often structured using a standard or scaled Lehman Formula. The second model is hourly: $150 to $350 or more an hour for a set number of hours a week, best for a buyer doing their own outreach but needing help on valuation, offer structuring, or negotiation.

Three places to find one: Axial, a deal-matching platform best for smaller transactions where a buyer posts exact criteria and gets private introductions; the IBBA (International Business Brokers Association) Directory, for finding credentialed individual brokers; and the BizBuySell Broker Directory, for finding active local brokers already networked in a specific region.

Outside your usual reading: a hobbyist's physics demo, not an AI story. A writer at maurycyz.com showed that an ordinary TV antenna, pointed at a normal broadcast signal, can detect a person moving through walls with no special equipment at all.

The trick is something called selective fading: a direct radio signal gets partly canceled out by a strong reflection, off a wall or a person, and because that cancellation depends on frequency, it does not look the same across the whole signal. He showed this with a spectrogram (a picture of how strong a signal is at each frequency over time). Standing still barely disturbed the pattern, but moving the antenna by just a few centimeters, or simply getting up from a chair, visibly changed it. Walking through a neighboring room produced a series of stripes, one each time his body moved far enough to flip the cancellation again.

He used the spacing between those stripes to work out his own walking speed. The stripes were about 70 pixels apart on a readout that updates 150 times a second, giving a gap of about 0.47 seconds between them, or a rate of about 2.1 cycles a second. Multiplying that by the signal's wavelength, worked out from the speed of light divided by the station's 490 MHz frequency, gave a walking speed of about 1.3 meters a second, or 4.6 kilometers an hour. Because he was walking straight across the signal's path, no extra correction for angle was needed.

The same trick works with WiFi or mobile signals, not just TV broadcasts, though those give a messier pattern that takes more work to read. The source here is thin on detail beyond the one demonstration, so there is not much more to add.