You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
I want to give AI an isolated IT environment, deliberately break things, and find out how much it can handle without me stepping in. I run technology for a small public school district. I rebuild networks, manage servers and devices, automate workflows, and build software when the tools we have don’t do what we need. A lot of IT work comes down to noticing something broke, figuring out why, fixing it, and checking that it actually works again. We already have tools that monitor systems and tools that can make changes. I want to see how well AI can connect those pieces and take responsibility for the result. I’ll build an isolated miniature enterprise with servers, networking, endpoints, identity, and monitoring. Then I’ll deliberately break things and give the AI a chance to diagnose and repair them. I don’t know how far it will get. That’s what I want to find out.
The bigger question is whether AI can run an IT department. The first version will test a smaller part of that: can it handle common infrastructure failures safely? I’ll start with ten repeatable problems, like a stopped service, a full disk, a bad configuration, or a broken dependency. Each test will have a known cause and a way to check whether the repair worked. The AI will have access to monitoring information and a limited set of approved actions. I’ll track whether it identifies the cause, fixes the problem, makes unnecessary changes, and recognizes when it needs help. I also want to test what happens after its first idea fails. Does it keep trying the same thing? Does it check its work? Can it roll back a change? Everything will run separately from my employer’s systems, using synthetic accounts and data. I plan to share the non-sensitive setup, test cases, code, and results so other people can reproduce the experiments.
I’m requesting $10,000 to build and test the first version.
My estimated budget is:
$3,000 for compute and virtualization hardware
$2,500 for local AI inference hardware
$1,500 for networking equipment and test devices
$1,000 for storage and recovery equipment
$1,000 for model access and cloud testing
$1,000 for cables, adapters, and replacement components
I’ll use refurbished equipment and free software where practical. Before buying anything, I’ll check what equipment I can already use. I’ll contribute my time and initially work on this alongside my job. The plan is to build the environment, run the experiments, and publish the results over six months.
It’s me, Thomas Miller, based in Maxwell, Texas.
As a Technology Director, I’ve rebuilt network infrastructure around Aruba, deployed centralized monitoring and management, automated device and account workflows, built imaging and device management systems, and created internal applications for staff and students.
I’m used to figuring out how to make something work with a small team and limited resources. My experience is in building and operating the infrastructure this project will test. The autonomous part is the experiment.
The AI might give a convincing explanation that’s wrong. It might make a change without checking whether it helped, or make the original problem worse. It might also need so much supervision that using a simple script would be easier. Those are useful results if I can show exactly where and why it failed. I’ll compare the AI with scripted recovery where that makes sense and document unsuccessful attempts. I want an honest answer about what works, what doesn’t, and which decisions still need a person. The lab will be isolated and resettable, with limits on what the AI can change.
$0 raised for this project so far