Reviews, issues that have no meaning. Trust issues in submissions. Fomo.

Excess verbiage in communication. Design docs, root cause debriefs etc all are clearly AI generated with little thought to whether they are helpful to the reader. They are certainly helpful to the writer as they can just offload it to the AI.

Its hard to discern the kernel of truth that the author is trying to communicate. So much of documentation is now AI generated and rarely does it help in understanding code bases or diagnosing issues in production.

This was my experience at MSFT for the past year or so.

Workslop [1] in seemingly every possible dimension: excessive wording in Slack, with entire messages clearly not written or fully understood by the person writing; Notion pages with the same pattern; vast piles of internal engineering documents such as RFCs, ADRs and AVDs that were clearly not only not written by the author but also not even reviewed, ultimately conveying no information, a negative signal-to-noise ratio; the diagram equivalent of the same - visualisations with numerous typos and the characteristic visual quirks of diffusion-generated text, also conveying no actual information.

Agentic systems being deployed and adding more work in places: Slack messages now being picked up by workflows and AI integrations that automatically create messages, tickets, 'action items'; alerts triggering additional agentic investigation adding noise instead of reducing it (If you couldn't figure out actionable alerts before I'm not sure how an agentic system is going to help with that problem).

The removal or avoidance of individual accountability or ownership in many places: developer contributions that aren't understood by the person submitting them; loss of ownership and accountibility in work ("I asked Claude to...", "I worked with Claude to...", "Claude decided to...", "I don't know, Claude did it..."); destruction of trust between colleagues.

[1] https://hbr.org/2025/09/ai-generated-workslop-is-destroying-...

Besides the obvious AI-Generated messages and emails that provide no real value to anybody. (just sending the prompt would be more efficient) We are still yet to adopt AI into our workflow.

Officially we are only allowed to use Microsoft Copilot, which is great if you do not work in IT but utterly useless for development.

Therefore, every developer uses their own AI model / subscription, and we have a varying amount of slop being shipped. This leads to some developers, who adhere to the rules, being accused of falling behind because they can obviously not keep up with the developers who just push slop into the review.