Humane raised $230M and shipped 10,000 units. A $600K Indiegogo is still selling e-paper phones.

Humane raised more than $230 million to build the Ai Pin. It planned on 100,000 units and shipped roughly 10,000. On February 28, 2025 the servers went dark and every Pin ever sold stopped working. HP paid $116 million for the team, the software platform and the patents, and nothing at all for the device.

The Minimal Company was founded in 2023 in Southern California. It ran an Indiegogo in March 2024, raised a little over $600,000 — about a quarter of one percent of Humane's capital — and started shipping that September. The phone is still on sale at $449.99.

One of those companies wrote new software and put it in a box. The other took software that was already free and sold you the box.

What $449.99 actually buys

A 4.3-inch grayscale e-paper screen at 800×600. A 35-key physical QWERTY with a quarter-millimetre of travel. 6GB/128GB, four days of battery, NFC, a headphone jack, microSD, a fingerprint sensor. And stock Android 14 with the full Play Store — every app you already have, installable, today.

Read that spec sheet again for what isn't in it. No operating system. No app platform. No sync service, no assistant, no store, no account system. The Minimal Company wrote approximately none of the software on the device it sells.

What it built is a screen that renders TikTok badly.

That is the product. Not a feature — a refusal, manufactured into glass, on a device that otherwise runs everything.

You cannot fix software with software

Screen Time and Digital Wellbeing have shipped on every phone for the better part of a decade, and the global numbers have not moved. DataReportal's April 2026 update has the average internet user at 33 hours and 13 minutes a week of online media, with 96.2% reaching it by phone.

The reason those tools fail is structural, not motivational. A software limit is enforced by the same device it restrains, and every one of them ships with an override that is one tap away at exactly the moment you least want to honour it. The referee works for the player.

E-paper has no override. The refresh rate is the policy. You cannot tap "15 more minutes" into a panel that physically cannot render a 60fps feed, and no future update will make it possible, because it was never a setting.

Hardware is the only layer where a constraint cannot be renegotiated at 11pm.

The slab market is shrinking. That is not evidence for any of this.

Counterpoint now expects 2026 shipments to fall 13.9% to 1.08 billion units — the worst annual decline on record and the lowest volume since 2013. Q1 was already down 6%.

Do not read that as vindication. Nobody quit the smartphone in 2026. DRAM and NAND got expensive, bills of materials rose 20–30%, and the damage landed hardest below $200 — on the cheapest phones, bought by the people with the least choice. It is a supply shock wearing the costume of a trend.

The honest line is the flat one. Shipments peaked at 1.47 billion in 2016 and have not come close since; 2024's rebound was 6.4% and 2025's was 1.0%. The general-purpose slab stopped being a growth story a decade ago. It stopped being where the differentiation lived at roughly the same time, and that is the part that matters.

The software went free, so the constraint became the scarce thing

For fifteen years hardware was the substrate and software captured the value. That has inverted, quietly, and the Minimal Phone is the cleanest illustration of it on sale.

If you want to build a personal device in 2026, you do not need to write an OS, a browser, a keyboard, a radio stack, a messaging client, a maps app or a store. Android is free. The Play Store is free. Open-weight models are free. An ODM in Shenzhen will build the board, and a crowdfunding page will finance the first run.

Everything that used to require a platform team is now a line item. What is left to sell — the only remaining differentiated component — is an opinion about what the device is not allowed to do, expressed in a material that cannot be patched.

The minimum viable audience for a purpose-built device collapsed along with the cost of building one. $600,000 clears it.

The market funded the exact opposite

Humane spent roughly $23,000 of investor capital for every Pin that reached a human being. Rabbit shipped more than 100,000 R1s on $30 million — and its founder conceded that about 5,000 people used one daily. It sold well once and was never used.

reMarkable is the control group. An e-paper tablet that deliberately cannot browse, cannot notify and cannot run apps: over a million units, a $1B valuation, roughly $300M in revenue with $31M of operating profit in 2021, on about $24 million of disclosed equity. A thousandth of Humane's capital per device.

The split is not about form factor, or timing, or taste. It is one line:

• Devices that invented new software and needed a new box died. Humane and Rabbit both asked you to move your life onto a platform with no apps, no ecosystem and no history, and to trust that a company two years old would keep the servers on. One of them proved that fear correct within a year.
• Devices that took software you already run and changed its shape shipped. reMarkable, the Steam Deck, the Minimal Phone. Nothing to migrate to. Nothing to learn. The novelty is entirely in the physical constraint, and the constraint is what you paid for.

Where this goes: the compute comes home

Canalys counted 8.8 million AI-capable PCs in Q2 2024, 14% of shipments. Gartner forecast 43% of all PCs in 2025. Counterpoint expects NPU-equipped machines to cross half of global shipments in 2026. Different methodologies, one direction: inference silicon is becoming standard equipment rather than a premium line.

That is the part that turns a niche into a category. For fifteen years a personal device was a thin client to somebody else's data centre, which is precisely why the box was interchangeable — the intelligence was never in it, and one slab of glass was as good as another for reaching a server.

When the model runs on the device, the device stops being a window. Software becomes personal in the literal sense: not multi-tenant. It holds your context, it does not leave, and nobody else's usage patterns shape it. And the moment the intelligence is local, the physical envelope around it — the screen, the input, the battery, what it refuses to display — is doing real work again instead of being a delivery mechanism.

The Minimal Phone is not that device. It is the early, legible version of the argument: that the shape of the hardware is once more a claim about how the software behaves.

What will get this wrong

The constraint you bought is the constraint you are stuck with. E-paper is genuinely bad at maps, at video, at anything moving, and it will be bad on the day you need it. That is not a bug in the pitch — it is the pitch — but the failure mode is real and it arrives at the worst moment.

A $449 second phone is a luxury solution to a discipline problem, and most people who buy one already own the thing they are trying to escape. Attrition in this category has never been measured honestly.

The macro is hostile to exactly this price band. Bills of materials are up 20–30% in a memory shortage that hits small-volume buyers hardest. A company that raised $600,000 has no hedge and no leverage with suppliers.

But note the one risk it does not carry. The Minimal Phone runs stock Android on the public Play Store. If the Minimal Company folds tomorrow, the phone in your pocket keeps working — you keep a slightly odd Android handset. Humane's went dark 90 days after the announcement, because the software was the company. Buying the box instead of the platform is not only cheaper to build. It is the version that survives its manufacturer.

What we take from this

TEGO does not make hardware. The mechanic is the same one we run into every week.

Every product we ship now competes against a feature list that is, functionally, free. Anyone can generate a dashboard. Anyone can wire up a CRM. The capability is not the moat and has not been for a while — which means the decision that carries weight is not what to add, it is what to refuse, and at which layer to enforce it.

We run the agency from a text thread. Not because SMS is a good interface — it is a terrible one — but because a 160-character channel cannot grow a settings page, cannot grow a nav bar, and cannot become another surface to maintain. The constraint is load-bearing. It is e-paper for an operations tool.

Our client dashboards ship with most modules off. Not fewer features. Fewer decisions, enforced somewhere the client cannot casually undo them at 11pm on a Tuesday.

The question worth asking about any product you are buying or building is no longer what can this do. It is: what will it refuse to do when I want it to bend, and is that refusal a setting or a material?

If you want a system that is opinionated on purpose, come talk to us.