Ansur

Repetitive design brand work needs taste, but once it learns what’s expected you can most definitely deliver client facing outputs. Taste doesn’t necessarily mean new every time, all it essentially means something subjectively unique to you that reflects ur brand and your way of work! And it’s not exactly similar to other people’s work. Design is just one example

@useansur agree on the repetitive slice, autonomy wins there. but you named taste as the bottleneck, and thats the part personalized software cant close alone. it can replay the taste it learned, not make the fresh call. the moment taste is live, youre back in the loop.

The last 3% is a different product. An agent can get 97% of the way there, but if a human still has to check every action, you haven’t really unlocked autonomy. It’s like a self-driving car that still needs someone behind the wheel. The real jump in value comes from governance: clear policies, permissions, audit trails, and humans stepping in only when something goes wrong. Better models matter. But trust infrastructure is what lets them actually run.

@phibrowser Repetitive tasks in legacy industry don’t need co-presence. Most of knowledge work is repetitive tasks, the bottleneck is taste and reliability, which can be fixed by hyper personalised software

@useansur the self-driving frame smuggles in the goal: human as exception-handler, step in only when it breaks. but a lot of work isnt that. the last 3% isnt an error to catch, its judgment threaded through the task. that wants co-presence, not autonomy plus an alarm.

AI agents can navigate ERP, CRM, and production system. That doesn’t mean they should be allowed to post to it unchecked. The gap between a capable demo and a trustworthy production system is an enforcement layer outside the model. Thats what @useansur is: useansur.com/capability-is-…