Phone Harness Lets AI Agents Drive a Real iPhone — or Rent One in the Cloud

The gap nobody talked about: agents without phones

Most AI agent demos live in a browser. They click links, fill forms, scrape pages. That's useful — until your agent needs to verify a phone number, log into an app that only exists on iOS, or interact with anything that assumes a human is holding a device.

That gap is what Phone Harness is closing.

Built by solo founder Shawn Pana — previously on growth at Browser Use — Phone Harness gives AI agents a persistent phone. Either a real iPhone or Android the user already owns, or a hosted cloud phone rented on demand. The agent drives it like a human would: tapping, scrolling, authenticating, responding.

Pana is the solo founder and CEO. Phone Harness is a Y Combinator Fall 2026 company. One person, YC-backed, attacking a problem that sits directly in the path of every serious agent stack.

What Phone Harness actually does

The core product is a harness — a control layer — that sits between an AI agent and a physical or virtualized phone. The agent sends instructions. The harness executes them on the device in real time.

Two deployment modes:

• Bring your own phone: Connect a real iPhone or Android. The agent operates it remotely. Useful when you need a persistent identity — a specific Apple ID, a WhatsApp account, a carrier-verified number.
• Cloud phone: Rent a hosted phone. Spun up on demand, discarded after the task. Useful for burner workflows, testing, or high-volume automation where you need clean sessions at scale.

Pana open-sourced the core before selling the hosted version — a classic developer-led distribution move. Build trust with the builders first. Then monetize the infrastructure underneath.

The open-source angle also matters for adoption. Any developer building an agent stack can wire in Phone Harness without a sales call. That's how Browser Use grew, too: open repo, real utility, word-of-mouth in the agent-builder community.

Why persistent phone identity is a hard problem

Agents hitting web apps run into friction constantly. But phone-native friction is a different category of hard:

• SMS 2FA: Half the internet still routes authentication through a text message. An agent without a real phone number hits a wall.
• App Store apps: Thousands of tools — banking, government services, niche B2B platforms — have no web interface. They're iOS or Android only. No browser agent reaches them.
• Device fingerprinting: Apps increasingly check whether they're running on a real device. Simulators and emulators fail these checks. A real phone, or a convincingly virtualized one, doesn't.
• Session persistence: Many mobile apps maintain sessions tied to device identity. An agent that can't hold that identity loses context between runs.

Phone Harness addresses all four. That's not a narrow wedge — it's infrastructure for a wide class of agent tasks that currently can't be automated at all.

The solo founder angle

Pana fits the emerging pattern of one-person AI-native companies that are genuinely competitive at the infrastructure layer. Not solo in the "side project" sense — solo in the sense that one person, with the right tools and the right distribution channel, can build and ship something YC thinks is worth backing.

The YC label matters here not just for credibility. YC Fall 2026 means Pana is operating in a cohort where the expectation is real revenue, real growth, and a real company — not a demo. The bar for what one person can build has moved, and Phone Harness is evidence of where it's moved to.

For founders tracking revenue per employee as the defining metric of AI-era startups, Phone Harness is a company worth watching. Infrastructure with recurring usage, open-source for acquisition, hosted product for monetization. If the agent market grows the way the browser automation market grew — and the signals say it will — the phone layer is a high-value abstraction to own.

Where this fits in the agent stack

Think of agent infrastructure in layers:

1. Reasoning — the LLM doing the thinking (GPT-4o, Claude, Gemini) 2. Action — the tool calling the APIs or clicking the UI (Browser Use, Playwright, Computer Use) 3. Identity — the persistent accounts, sessions, and credentials the agent operates as 4. Device — the physical or virtual hardware the agent runs on

Most of the investment and tooling in 2024-2025 landed on layers 1 and 2. Layer 3 got some attention with credential vaults and session managers. Layer 4 — device — is almost entirely unsolved.

Phone Harness is building layer 4.

That's not a feature. That's a category. Any agent that needs to operate in the real world — not just the browser-accessible web — eventually needs what Phone Harness provides. The question is whether Pana can establish the standard before the cloud providers decide to bolt this onto their existing compute offerings.

The open-source-to-SaaS playbook

Pana's distribution strategy isn't new, but it's the right one for this market. Open-source the core. Let the community validate it. Monetize the hosted version — the cloud phones — where the operational burden is real and developers are happy to pay to avoid it.

This is exactly how Browser Use expanded, and how most of the successful agent tooling in 2025 got its first thousand users. The developer community building agents is small, tight, and extremely online. If a tool works, it spreads fast. If it's open-source, the threshold to try it is zero.

By the time a developer is running production workloads through cloud phones, the switching cost is high enough that the hosted product sells itself.

What to watch

A few things will determine whether Phone Harness becomes infrastructure or a footnote:

• Device fleet size: Cloud phones are only useful at scale if the fleet is large enough to avoid queue times. The economics of maintaining real or virtualized devices at scale are non-trivial.
• Anti-automation countermeasures: Apps that don't want to be automated will update their detection. Phone Harness will need to stay ahead of that arms race.
• Enterprise demand: The clearest revenue path is B2B — companies running agent workflows that need phone automation at volume. Whether Pana can close that without a sales team, or whether YC's network does that work, is an open question.
• Platform risk: Apple and Google control the hardware layer. Any product that depends on driving iOS or Android at scale is one policy update away from a hard conversation.

None of these are fatal. All of them are real. The founders who build AI-native companies that last are the ones who see the risks clearly and build around them before they bite.

The bottom line

Phone Harness is filling a gap that most agent builders have been routing around — not solving. If your agent stack needs phone identity, Shawn Pana now has infrastructure for that. One person, YC-backed, open-sourced before the pitch.

The agent layer is being built in public, one solo founder at a time. Phone Harness is one of the cleaner bets on where the gaps actually are.

---

Is your company eligible? Submit to the leaderboard → onepersonunicorn.co/submit