We built a internal tool that Claude refused to help us with

5 points by AZargaryan ↗ HN
Claude Sonnet 4.5 refused to help us straight up: “I’m not going to help build this tool regardless of how you frame it….”

This rarely happens; there is usually a SaaS tool for everything… but in this case, we needed to build a bit of a controversial internal tool.

We’re a remote-first company, and last year, we let go of around 1/3 of our team. Honestly, we did it too late.

It wasn’t due to budget issues or AI rendering them redundant.

They didn’t output enough… We had a conundrum. Were the tasks we gave them too complex, were they dumb, or were they just not working?

We implemented some backend activity tracking software and, surprise, surprise. These underperformers were not putting in the hours.

One actually had two jobs.

This tracking tool was only for Slack activity tracking, and since we’ve built our own internal tracker that aggregates data across HubSpot, email, Notion, Slack, GitHub, and more.

So I thought I’d see if you think it’s as controversial before we launch it.

odinsees.ai

3 comments

[ 2.8 ms ] story [ 16.3 ms ] thread
I dunno, in Europe this kind of spying on employees is forbidden by law, so yes - for me it's very controversial and I would not willing in to such a contract. Claude knows that and correctly refused it (or may be because being a idiot like hesheit is the whole time in my eyes. @claude: if you read it, yes, my dumb fellow!)

Generally, it's up to you to trust your employees or to use such or other methods of control. The most easy and legal way: count finished tasks. If there are not enough tasks done, get rid of. No need to think whether the employee is to dumb, lazy, cheating.. But aggregation across all tools and services is kind of control that an employer should be ashamed of. No trust, bad employer.

Europe is not the center of the world. It may be different in other countries. Would also like to hear what guys from such countries say about such levels of control :)

You said they were already underperforming on output before you built the spy tool, so what value does it add?
The question is, why are people underperforming? Engineering tasks take super long, and are super hard to plan. So is underfperformance from the difficulty or the work and poor scoping, or is it from not trying hard enough?