If you want to run Codex, Hermes, or other AI agents for extended periods, you don’t necessarily need to buy the latest Mac. This video tests a 2017 MacBook Pro with its faulty screen removed—nicknamed the “headless knight”—to see whether a nine-year-old Intel Mac can serve as a low-cost AI agent host, and what key differences exist between it and Apple Silicon Macs.
Video chapters
- 00:00 Low-cost AI Agent host sharing
- 00:45 Why recommend a headless MacBook
- 01:20 What is a headless MacBook
- 02:10 Why headless MacBooks are popular again
- 02:55 2017 MacBook Pro test unit specifications
- 04:20 Actual Usage and the AI Agent Environment
- 05:40 Questions you need to know before buying
- 06:20 Upgrading macOS using OpenCore
- 07:10 Claude Code, Codex, and Hermes Compatibility
- 08:35 Limitations of Running AI Agents on Intel Mac
- 09:55 Actual usage experience
- 11:20 Is a headless MacBook worth buying?
- 12:35 Summary and Buying Recommendations
What is a “headless” MacBook?
“Headless Horseman” refers to a MacBook whose screen is damaged or has been removed, but the rest of the machine still works. Since repairing a MacBook screen can cost nearly as much as a whole used unit, some machines are bought back at low prices and have their screens removed, then paired with an external monitor, keyboard, and mouse for use as desktop computers.
This type of machine doesn’t have a built-in display, which is inconvenient for typical laptop users. But since AI agent hosts are primarily used for long-running connections, automation, and remote operations, they don’t really need to rely on a built-in screen. The MacBook still retains its battery, keyboard, and trackpad, so it won’t shut down immediately during a sudden power outage—a distinct advantage over older desktop machines.
2017 Intel MacBook Pro specs and real-world usage
The test unit in the video is a 2017 MacBook Pro with a 7th-generation Intel Core i7 processor, 16GB of RAM, and a 512GB SSD, equipped with a Touch Bar, discrete graphics, and four Thunderbolt 3/USB-C ports. By connecting an external display, keyboard, and mouse through a USB hub, it can be used just like a regular desktop computer.
The test environment is set up with Docker, Codex, Hermes, and other commonly used programs, ensuring smooth operation for everyday web browsing, video playback, light gaming, and cloud-based AI Agent use. The machine shown in the video can run for about three hours on battery power, but since the battery health, keyboard, trackpad, and port conditions vary greatly among used units, the results from a single test unit should not be taken as a guarantee for all machines.
The First Hurdle for Older Intel Macs: macOS Version
Older Intel MacBooks are officially limited in the macOS versions they can support, and some AI tools, Homebrew, browser automation packages, or plugins require macOS 13 or even macOS 14 and above. The machine in the video was therefore upgraded to macOS 14 Sonoma via OpenCore Legacy Patcher; for the actual steps, refer to the link in the video description. OpenCore Legacy Patcher Upgrade Tutorial。
Such unofficial upgrades may encounter driver, update, performance, or reinstallation issues, and are not suitable for those who rely on their computer as their only work machine. Before purchasing, you should also confirm whether the seller allows system reinstallation, how the current system is installed, and whether you can handle boot or operating system problems yourself in the future.
Can Codex, Hermes, and cloud AI Agent be used?
If the main tasks are calling cloud models, writing code, searching for information, modifying files, or running web automation, Intel Macs still have practical value. The video tests the CLI/Web UI workflows of Codex and Hermes, and there is no noticeable difference in daily responses compared to other hosts; operations like interacting with web page elements, uploading content, or clicking buttons through the browser also work fine.
However, some desktop applications, helper programs, or installation packages may not offer Intel versions, and whether they can be installed may change with software updates. Before you make a purchase, first list the Agents, desktop programs, and plug-ins you rely on, and confirm the processor architecture and minimum macOS requirements for each one.
The gap between Intel Macs and Apple Silicon in running AI.
The main limitations of Intel Macs show up in local computing and full desktop control. For local large language models, speech transcription, OCR, image recognition, or other machine learning tasks that benefit from Apple Silicon optimization, M-series processors generally deliver better performance and software support. The video also notes that some Computer Use or desktop features requiring native helpers may not be fully available on Intel models.
If the work only involves sending requests to cloud models while running programs, files, and browser tasks locally, the processor gap won’t necessarily show up directly in every operation. But for long-term use of new features, running multiple Agents at once, or handling models and multimedia data locally, Apple Silicon paired with a newer macOS would be the safer choice.
Is a headless MacBook worth buying?
These machines are suitable for people who can get them at a very low price, understand the risks of secondhand equipment, and are willing to handle unofficial system upgrades themselves. If you only need a backup host for running cloud-based AI agents, automation scripts, and light workloads, an Intel Mac with 16GB of memory can still provide value.
However, headless MacBooks usually have no warranty and may also have issues such as sticking keys, unstable ports, battery degradation, or other undisclosed problems. If the price is already close to that of a used M-series Mac mini or MacBook, it would be more reasonable to go with Apple Silicon directly, as it also offers longer system and software support. The video’s conclusion is not to encourage impulse buying, but rather to explain that under the premise of a sufficiently low price and clear use case, an older Intel Mac can still serve as a usable small AI agent host.
Source: KOCPC Chinese