Great advice regarding the middle ground. My experience with teams using AI is that the adoption levels are on the lower side - the majority do not know what they can do with the more powerful models (other than research), and the "extreme adoptors" are a very small group who are pushing the limits. At CE it sounds like your policies are a "one size". Do have to impliment different levels of access (or abilities, such as with creating agents) for some users over others, and how do you manage that? Great article and something that is not spoken about very often - the real implimentations in finding that middle ground.
Thanks! Yeah I think we should do better at implementing different restrictions for different folks, and in fact we're exploring that now via different Iru blueprints for different types of staff.
When you say: "we sandbox Claude Cowork’s network access and only let it access a select list of approved websites. It can still use the web_search tool to get information from the internet"
Can you give examples of what is blocked under this policy? Is the approved websites function just for which websites it can, e.g. download things from?
Hi! Thanks. We actually block all websites except coefficientgiving.org, as some folks wanted to use cowork to scrape some data from our site. We also check the 'package maintainers' allowlist checkbox, so Cowork can install new packages if needs to. Does that answer your question?
Great advice regarding the middle ground. My experience with teams using AI is that the adoption levels are on the lower side - the majority do not know what they can do with the more powerful models (other than research), and the "extreme adoptors" are a very small group who are pushing the limits. At CE it sounds like your policies are a "one size". Do have to impliment different levels of access (or abilities, such as with creating agents) for some users over others, and how do you manage that? Great article and something that is not spoken about very often - the real implimentations in finding that middle ground.
Thanks! Yeah I think we should do better at implementing different restrictions for different folks, and in fact we're exploring that now via different Iru blueprints for different types of staff.
Super useful!
When you say: "we sandbox Claude Cowork’s network access and only let it access a select list of approved websites. It can still use the web_search tool to get information from the internet"
Can you give examples of what is blocked under this policy? Is the approved websites function just for which websites it can, e.g. download things from?
Hi! Thanks. We actually block all websites except coefficientgiving.org, as some folks wanted to use cowork to scrape some data from our site. We also check the 'package maintainers' allowlist checkbox, so Cowork can install new packages if needs to. Does that answer your question?