This makes me wonder: what do developers, who completely rely on LLMs to write their code, do when the service is down?
I realize this is already a problem for other jobs, which require working with SAAS, but it seems odd to me that now some developers will fall into this "helpless" category as well.
Early in my career (which started in civil engineering) I was working with a man at the very end of his, which started in the 1950s. I was the young tech-focused intern who found a way to use a computer for everything even when printed and sometimes hand-drawn plans were the standard of the day. He asked me once if I knew how to use a slide rule, which I didn't.
"Well, what do you do when the power goes out?", he asked.
"I go home, just like you would.", I said with a smile.
He paused for a moment and nodded, "you know, you're absolutely right".
Nice story. I guess it can be looked at as some sort of parable. But if I take it literally: I never had a power outage at work, but SAAS downtime happens every year (probably multiple times).
Make sure you have enough vRAM around for the local models!
or, take the brief, unscheduled break.
Most corporate cultures need it ;]
Non-serious answer: I'll go for a walk.
Serious answer: I can write code manually, but it feels like a waste of time. I'll just go for a walk to synthesize my ideas if a service was down, and I don't think not writing actual code for a day is a huge problem. So focus on health and maybe even talk to humans.
Doubtful all services are ever down simultaneously, either way:
What do devs do when Github or Gitlab is down?
AWS? GCP? Azure?
Or whatever Atlassian product they're using.
Plus, most devs do a bit more than just produce lines of code.
> This makes me wonder: what do developers, who completely rely on LLMs to write their code, do when the service is down?
Even the engineers at these AI companies can't use these LLMs to fix an outage when there is one. Especially SREs.
But if one has to just sit there and "wait" for the outage to subside then perhaps the kitchen timer just went off and declared that these "developers" are cooked.
As a technical ... I don't know what to call myself anymore; not a product manager but not a developer nor engineer.
The answer: audibly swear out loud.
Claude/codex herder.
They’re like sheep, but a bit smarter.
I'm calling myself Vibe Director.
Wetware interface. You are the wetware interface between the company and the AI it uses.
Even worse the on-call folks who need to do recovery but need an LLM to do it, but surprise they are both on the same cloud provider.
If they're smart, they just switch to ppq.ai or an openrouter provider where they can purchase prepaid tokens from various providers with many alternative models available.
Switch to codex? :)
Exactly, these agentic coding tools which operate on your codebase present on your own disk are a lot more fungible than most SaaS.
Use Qwen3/other locally via lm-studio.
Same as if GitHub were down over the last decade.
Prepare the prompts for the features.
What do most people do, when the power goes out?
We usually try to figure out how to build reliability/redundancy in step with what we require to function as a society under most circumstances without taking outsized losses.
When things go worse than anticipated, we take the hit, try to recover and maybe learn to strengthen the system afterwards. I would rate us roughly okay-ish at that, mostly because I don't know what to compare it to, since we are the only species to do it at this level to my knowledge.
Grab a coffee and wait.
Use another AI tool :D
I mean, same thing when other web services are down. Try updating your packages without a web connection...
Switch to Grok or vice versa (I have Pro on both), or use any one of the other free tier LLM's (Gemini, ChatGPT etc)