Photon's $4.5M bet on texting agents rests mostly on open-source use

Daniel Tian's first product to spread was a bot that texted his friends back as him over iMessage ^[11]. Before that, he and Ryan Zhu, now Photon's CTO, built consumer apps at student hackathons and ran into the same problem every time: getting people to discover them ^[10]. The bot did not have to be discovered. It worked inside Apple's Messages, an app the people it reached already used every day ^[12]. When the pair open-sourced it on GitHub, thousands of developers adopted it ^[11].

Photon's thesis grew out of that bot. As TechCrunch describes it, users no longer want to install new apps and want agents that work through the messaging apps they already use ^[6]. Tian expects a slow transition. "Honestly, it's going to take a while, but I think, directionally, it's inevitable," he said ^[5].

Product teams like to repeat the first half of that thesis: nobody wants another app. Here's what users actually did this year. They pushed Meta's Muse, a standalone app, to No. 1 on the app stores ^[7]. Tian makes a narrower claim. Even if Muse comes to dominate consumer AI, he sees a market among developers who want their own applications to work over messaging ^[8].

Photon's own figures mostly measure developers. The 40,000 figure counts sign-ups, according to the company ^[2]. By Photon's account, the open-source version still accounts for 98% of use ^[13]. Roughly 2% runs on the hosted platform that carries its subscription tiers ^[1]. Photon says revenue grew 10x in four months ^[3] but is not sharing revenue figures ^[17]. Messaging volume grew 5x in the last month, the company says ^[19]. More messages can mean more users, or more back-and-forth to finish one task. The closest thing to a retention number is churn, which Photon puts below 3% ^[18]. Tian estimated that across both versions the technology reaches millions of end users ^[23].

The customer list shows who this is for today: Corgi Insurance, business banking platform Rho, Gen Z dating platform Ditto, social introductions app Boardy, finance assistant Fliptexts and AI email client Slashy ^[20]. Several of those are jobs people already handle in a text thread, like asking about a policy or getting introduced to someone.

I'd plan for messaging agents as a second channel for conversational jobs and leave the app roadmap alone for now. The tradeoff is that the evidence so far comes mostly from developers trying the tool, and the end-user case rests on Tian's estimate and one churn figure.

Two questions sort the decision. One is whether the job fits in a text thread, like a question or a confirmation, or needs a screen for comparing options or editing. The other is whether the team's problem is getting found or keeping people. A text-sized job with a discovery problem is where Photon's founders started, and a messaging agent is the cheapest first test. If the job is text-sized but users drift away, run an agent alongside the existing app and judge it on repeat use. A job that needs a screen and suffers from discovery needs an acquisition plan, because a chat thread cannot show a comparison table. When the job needs a screen and users leave anyway, the work belongs in the app. In each quadrant, the numbers to collect are time to the first completed task and the share of users who come back to complete another one after 30 days.