# document-based work becomes obsolete, as human-first systems of record why can't you use a chatbot to compile a project plan from 5 documents, 20 emails, and your own personal notes today? well, you can, but the results are generic and boring crap. there's no high-quality, professional tool that would do that -- although this is very much in the realm of capability as of october 2025. i find it weird how little useful work is being addressed by ai startups, building a whole new saas just to analyze your contracts. excuse me, but those things should not be separate tools, and separate companies built around them, and charging for them. if i'm pushing my documents around in sharepoint, or sap, i gotta be able to push them faster with ai. or better yet, collapse the whole pushing thing into one prompt (in company vernacular, it's called a business process). if we take a step back, why can we collapse processes? at heart, they are nothing else but changing state and form of structured documents (think orders, invoices, tickets, ... ) based on certain events (order creation, confirmation, planning, ... ) and evaluation criteria (material available, machine compatible, delivery date selected, ... ). and language models are great at understanding & generating structured text, and can apply predefined criteria quite well. what follows? one by one, most work on document creation & management will become automated. artists suffer because songs, paintings, photos are documents that do not depend on events and evaluations, they just exist -- and are easily automatable. orders or invoices, while much simpler in their dasein, need to be connected to information inputs from other organizations, i.e., have way too many dependencies, and must be tracked by dozens of transactional entries in systems of records, to be valid and actionable. but they too will be automatable, as soon as org2org communication becomes a thing. ** in the next step, the documents themselves will become obsolete because no AI needs a PDF of a purchase order to confirm it and start production. OCR technology will stop making sense in the enterprise setting because there will be nothing to scan. layout approach of OCR is outdated anyway as it is vastly surpassed by the flexibility of vision models and their flexibility. in the course of deprecation of document-based systems, many incumbents will go extinct (looking at you, sap, dynamics, oracle erp, salesforce -- name a saas to fail in the comments!). AI-first products will emerge, which means software products created for an AI as the primary user, not for humans, with [[ax]] in mind. imagine an ERP system built that way from the ground up, for an AI-only user & AI-only workloads. a lot of the use cases that people are excited about now, including those that i sell to my customers, will become obsolete -- just like most of the workforce. --- // 26 oct 2025, berlin