Julian Oliver6d ago

Yesterday an old friend asked me of the impact AI is having on my #sysadmin work, whether I was using it. I responded that it was not, & that using it would be unproductive as I'd lose so much time auditing deployments by an agent I cannot trust for slop, bloat & flaws.

What I didn't have time to add was that I already have v low carbon, 100% sovereign automation I can trust: shell scripts. I know exactly what they do, when & why, because I wrote them. Such value & confidence is irreplaceable.

Show threadJulian Oliver6d ago
(Yes, there's also Ansible etc. I'm old school)
Etoneya5d ago
Show threadJulian Oliver

Furthermore, I could not, in any semblance of good faith, hand over server infrastructure to an organisation (esp one under severe adversarial pressure) that has been deployed with the assistance of an 'agentic' AI. This would be completely unethical because I could not speak from the infrastructure with knowledge, that it was capable and safe to house them and their work.

Trust comes from knowledge. There is no knowing a black box.

Jun 8 at 7:42pmWeb
Show threadTom6d ago
@JulianOliver There really is no decent substitute for a properly written shell script that does one thing well.
Show threadChris L6d ago
@JulianOliver I would have to say the same for v1 of services I have written. That is why I use methods such as harder than prod testing, gradual ramp, shadow mode, and so on. Trust comes from knowledge but especially knowledge that the code works in the environment at hand. But it is rare I look to the compiler or firmware as being something to understand in depth. So some black boxes are necessary. 1/2
Show threadChris L6d ago
To argue against AI, I focus on the folly of throwing trillions of dollars at something with the current capabilities, especially with some urgent needs competing for those resources. Even for coding (which is a sort of best case of lots of public training data and rigorous frameworks to validate output) it's not making my skills useless. /2