# total automation of organizations: the task, the path, the seed, the loop everybody believes that work will be automated, so let's have a look at the mechanism. first, we automate granular, well-scoped tasks with the human in the loop -- it's happening now with chatbot tools, workflows, and other wrappers -- that's 1/ slowly but surely, we will automate more and more activities in business processes until we reach end-to-end automation in many of them, esp. those that are purely document-based and to not require any physical interaction. think usual suspects like customer service, but also a lot of legal, marketing, admin. while the work of an insurance appraiser, while akin to customer service, is more robust, as for many cases, on-site presence is irreplaceable with smartphones photos of the insured -- now that is 2/ next, after most processes are end-to-end automated, a logical thing to do is to remove the human from the loop altogether -- by closing in the automation on itself, and treating organizations as running on loops, not processes. when marketing, product development, invoicing, and support are 99% ai-driven and have all worked flawlessly for 2 years, what would a business owner do? in parallel, the trend of AI-aided communication will continue, emerging as AI-first products will require AI-first communication protocols. in 20 years we will reach the state of total automation, when there's pretty much no piece of work focused on document creation left to humans. it's laughable when AI CEOs say it's gonna happen in 2 years -- that tells me they have never dealt with a corporate IT department. however, our brave heroes in IT administration, despite all their efforts to slow down every conceivable project, will fail. on the other hand, of course, the [metr study](https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/) & [reports on sonnet45](https://www.reuters.com/business/retail-consumer/anthropic-launches-claude-45-touts-better-abilities-targets-business-customers-2025-09-29/) suggests that ai can run for up to 30h already, but it depends much on what it performs during those runs. total automation is inevitable -- because it is doable, and makes economic sense. --- // 26 oct 2025, berlin