# Blue: an agent that runs tasks on your iPhone

Canonical: https://mudpie.ai/companies/blue/
Breadcrumb: [Home](https://mudpie.ai/) / [Companies](https://mudpie.ai/companies/) / [Blue: an agent that runs tasks on your iPhone](https://mudpie.ai/companies/blue/)
Author: Ali Abouelatta (https://mudpie.ai/authors/ali-abouelatta/)
Published: 2026-09-19
Updated: 2026-09-19
Research type: Company profile
Method: Company and accelerator sources checked 2026-09-19. Product claims are attributed to their sources; this is research, not a hands-on product trial.

## What it does

Blue is a small USB-C dongle and iPhone app that lets a person ask for work to be done inside the apps already on the phone. The [YC profile](https://www.ycombinator.com/companies/blue) describes tasks such as sending a follow-up, booking an appointment or answering a message, with Blue opening the relevant app and reporting back without requiring the user to touch the phone. The current [homepage](https://heyblue.com/) is sparse and says only that the company is building something that does the next task for you.

The fit is a consumer who has repetitive phone workflows and is comfortable delegating device control to a dedicated hardware/software assistant. The product’s promise is broader than dictation or a voice shortcut: Blue is meant to navigate existing apps and complete actions. That makes the permission, confirmation and failure model the core of the product.

## Why I’d look closer

The team combines the right kinds of experience for the hardware/software bet. The [YC profile](https://www.ycombinator.com/companies/blue) describes Omar Abdelaziz as a former Google Assistant and Scale AI leader with an early LLM computer-use patent, Peter Krogh as a long-time voice-dialog and Gemini Live design lead, and Farhan Hossain as an Apple Vision Pro engineer with robotics, wearables and consumer-electronics experience. Those are public professional backgrounds, not evidence that Blue can safely operate every iPhone app.

The product history also matters. The profile includes an earlier Blue launch for a smart dictation keyboard, then a previous launch for a voice assistant, before the current “run your phone” description. That looks like an evolution toward computer use rather than a reason to merge every past feature into today’s product. No current public pricing, app-store proof or integration list was exposed in the sources checked.

## What I’d ask

Which actions can Blue perform without confirmation, and how does a user see or undo a step? What happens with two-factor prompts, private messages, payments, destructive actions or an app redesign? I’d test a low-risk appointment workflow, inspect permissions and logs, and require an explicit approval boundary before delegating anything irreversible.

## My editorial take

Blue is interesting because the hardware could give computer-use agents a physical, always-available interface. The buyer question is not whether voice can launch an app. It is whether Blue can complete a multi-step task reliably while keeping the user in control. The current sparse homepage means this should stay a sourced early profile, not a product recommendation.

## Quick facts

| Field | Sourced detail |
|---|---|
| Buyer fit | Consumers with repetitive, multi-step iPhone tasks |
| Product wedge | USB-C dongle plus iPhone app for agentic app actions |
| Founder context | Google Assistant, Scale AI, Gemini Live and Apple Vision Pro backgrounds |
| Public pricing | Not exposed in the sources checked |

## Sources checked

Checked 2026-09-19.

| Source | Used for |
|---|---|
| [YC company profile](https://www.ycombinator.com/companies/blue) | Product, founders and launch history |
| [Blue homepage](https://heyblue.com/) | Current public status and sparse product surface |

## Cohort context

Blue is listed in Summer 2025. In our 2026-09-18 directory snapshot, 12 of 166 listed companies in that cohort have YC’s primary industry label Consumer (7.2%). This is a current-directory comparison, not an original intake count or a performance ranking. [Nine-cohort dataset](https://mudpie.ai/research/yc-cohorts-2026-09-19.json).

## Public website snapshot

Observed 2026-09-19T16:17:58.187Z in raw homepage HTML. This records visible metadata and advertised links, not agent execution or product quality.

| Signal | Homepage observation |
| --- | --- |
| Product description metadata | Observed |
| Canonical link | Observed |
| H1 or H2 heading | Observed |
| Typed structured data | Not observed in this response |
| Docs/developer link | Not observed in this response |
| Pricing link | Not observed in this response |
| llms.txt link | Not observed in this response |
| Markdown alternate | Not observed in this response |

[Public observations](https://mudpie.ai/research/yc-homepage-links-2026-09-19.json) · [Collection method](https://mudpie.ai/research/yc-homepage-methods/README.md). Missing links here do not establish that a capability or file is absent elsewhere.


## Author disclosure

I cofound Lazyweb and publish Mudpie. This is an owner-written publication, not an independent testing organization. Research notes distinguish observations, sourced reporting and editorial judgment.