What Is an AI Shed? The Always-On Agent Machine Idea Explained
Practitioner notes on AI, SEO, PPC, and agency operations.
In early October, DHH posted one sentence that spread fast through the developer world:
Every developer needs an AI shed: an always-on machine running on their tailscale network where the majority of their herdr agents are running.
Source: DHH on X, October 4 2026.
If you run a small business rather than a software team, you are probably wondering why this matters to you. It is a fair question, and the honest answer is that it is not about building a shed.
It is about a small shift in how work happens when AI is part of your operation. This post explains what an AI shed is, what it costs, what it does not solve, and how to tell whether any of it is worth your time.
First, what "AI shed" actually means
The term is DHH's, and it started as a developer joke that turned into a real pattern.
An AI shed is an always-on computer on your own private network where most of your AI agents live and keep working. Your laptop can close. The work does not stop.
The practical details matter here, because the difference between an AI shed and ordinary cloud usage is not the AI. It is where the AI runs and whether it keeps running when you stop looking at it.
DHH runs several of these small machines together. His main one is a Minisforum MS-A2, and he says a Beelink box with an 8745HS handles most of the work just as well. In a later interview he described pulling four mini PCs out of a closet, hanging them off KVM adapters and a WireGuard network, and running roughly sixteen threads at once. He is explicit that the human is the bottleneck, not the hardware. Source: DHH on the shed hardware.
The session runner he uses is called Herdr. With it, you can disconnect from a running session and leave the agents working. When you come back, the session is still there.
Why it suddenly matters
Here is the change that made the idea resonate.
For most of the AI era, your work has lived on your laptop. You open a chat, you prompt, you wait, you copy the result, you close the tab. The work exists only while you are sitting there watching it.
That is a strange constraint for business work, which mostly happens while you sleep.
An AI shed removes the constraint. If you want content drafted, a site audited, or reporting compiled, that work can be handed to an agent and left alone. You come back in the morning to finished output instead of a running task.
DHH framed it as developer infrastructure. For a business owner, read it as batch work that survives the night.
What it would look like for a small business
Let us be concrete, because the concept sounds heavier than it is.
You would get a small box, a few hundred dollars, put it somewhere out of the way, and plug it in. It stays on. You connect to it from your phone or laptop over a private network such as Tailscale, so you can reach it from anywhere without opening a port to the internet.
Then you hand it work: monitor a set of pages for changes, compile a weekly report, draft content from a list you approved in the morning, check whether your listings are still consistent.
The machine keeps going while you are at a customer appointment, in a meeting, or asleep.
How the work flows
That is the entire concept. The word "shed" is about where it physically sits, and the reason it needs a home at all is privacy, which we will get to.
What it costs
Less than people assume.
- Hardware: a few hundred dollars for a competent mini PC. DHH explicitly says you do not need an expensive machine for this.
- Power and space: a closet shelf and a few dollars a month of electricity.
- Ongoing cost: DHH notes his setups are mostly API calls, meaning you still pay model providers for what your agents actually do. A higher-end machine could hold local models, which changes that math.
There is a real learning curve, and it is not a weekend project for most people. The hardware is the easy part. Running things reliably on a machine nobody is watching is the work. If you want that built for you rather than built by you, that is the same work we do as AI services and agent development.
What it does not solve
This is the section most posts about AI hardware leave out.
An always-on machine does not think for you. It will cheerfully produce a large volume of plausible work that still needs a human decision before any of it reaches a customer.
If you do not already know what good output looks like in your business, an agent working overnight will not teach you. It will just generate more work for you to review.
There is also a maintenance reality. A machine nobody watches eventually stops. Updates, failed tasks, and disk space all need attention. Somebody still owns the outcome.
The shed runs the work. It does not own the judgment.
The privacy angle is the real argument
This is the part that matters most if you work with other people's data.
Running agents on hardware you own means your code, your files, and your API keys stay in the building. That is not a theoretical benefit. Plenty of people work under contracts that forbid moving client material off a company machine, and cloud tooling can violate that without anyone intending to.
If you are a consultant or an agency, that constraint is not a detail. It is the whole ballgame.
For a business owner handling your own data, the argument is simpler: you can see what runs, where it runs, and what it touches.
The counterargument
It is worth hearing the other side.
David Ondrej pushed back that at scale, this stops being enough. His argument is that you will eventually run more agents than one home machine can hold, and each one wants its own isolated environment. His conclusion: having an AI shed is a short-term fix, and long term you go to the cloud. Source: David Ondrej on X.
That is a reasonable position for someone running serious infrastructure. For a small business running a handful of recurring jobs, it is not where you are headed yet.
| AI shed | Cloud | |
|---|---|---|
| You get | Privacy, ownership, predictable cost | Scale and reliability |
| You give up | Uptime and maintenance you now own | Less control over where your data sits |
| Best fit | A handful of recurring jobs | Isolated workloads at scale |
The practical read: an AI shed makes sense when your work is repetitive, mostly text, and comfortable running on one machine you control. It stops making sense when you need isolation at scale, guaranteed uptime, or elastic capacity you would rather not own.
Do you need one?
Honestly, most small businesses do not need to build anything.
Most of what an AI shed does, a well-run monthly retainer already does for you, with more judgment and fewer hardware problems.
It starts making sense when you have recurring work that is genuinely tedious, you already pay for AI tooling, and you are comfortable owning a small machine and its upkeep. Or when data privacy is a hard requirement rather than a preference.
If you are still deciding whether fundamentals are covered, get those right first. Rankings, content that answers real questions, and consistent business listings do more than any automation layer.
Questions we get asked
What is an AI shed?
An AI shed is an always-on computer on your own private network where most of your AI agents run, so your work keeps going after you close your laptop. The term comes from DHH, and the name refers to where the machine physically sits.
Do I need a physical machine to do this?
No, and most small businesses should not build one. The same work can be done through managed tools. An AI shed only earns its place when you have repetitive work, already pay for AI tooling, and are comfortable owning a machine and its upkeep, or when data privacy is a hard requirement.
How much does an AI shed cost?
Expect a few hundred dollars of hardware plus a few dollars a month of electricity. Most setups are API calls, so you still pay your model provider for what the agents do. A higher-end machine could run local models instead.
Is an AI shed better than cloud hosting?
Not inherently. It is a different tradeoff. A shed gives you privacy, ownership, and predictable cost, at the price of uptime and maintenance you now own. Cloud gives you scale and reliability, at the price of less control over where your data sits. For a handful of recurring jobs, an owned machine is reasonable. For isolated workloads at scale, cloud wins.
Will an AI shed replace an SEO agency?
No. It runs repetitive tasks. The judgment about what is worth doing, whether the output is right, and whether it actually helped your business still requires someone who understands the outcome. Automation reduces the cost of the mechanical work. It does not replace accountability.
The bottom line
Key takeaway
An AI shed is a small always-on machine where AI agents keep working after you close your laptop.
For developers it is infrastructure. For a small business it is the idea of work that finishes while you sleep, on hardware you own, using data that never leaves the building.
It is real, it is cheap by infrastructure standards, and it is not magic. It runs the repetitive work. The judgment, the review, and the accountability stay with you.
Which is, in the end, true of most good business tools.
