Whenever I am trying to work on a project, I go through it myself first with proper details that what I need to build and all the specifications then tell LLMs the plan so they can code and I can keep my architecture skills intact. For keeping my coding skills intact, for small code changes, I prefer doing them myself, so I don't make myself much dumb.
I usually make tab groups to sort the frequently visited tabs properly but still most of the times, I am opening a lot of tabs outside the groups then I keep closing all the tabs outside the groups.
Even if the LLMs are able to do managerial work, still they will need supervisors on top of them. According to me, it will more like team leads and managers with LLMs as their developers or maybe planners.
Another thing, it also depends on LLM costs as a lot of companies are exceeding their budgets and with more dependency on LLMs, the costs will be even more which can eventually reduce the company profits.
To be relevant in the software industry, we need to get upgraded all the time similar to an app. According to me, with the more development in AI side, it will be more like managerial job instead of proper dev job. You will be managing your agents, assigning them tasks so your technical knowledge should be up to the par so you can review them properly and keep yourself relevant. The development will be quick but still needs to be manually tested and manually approved like it is right now because the models will never be deterministic so they will require reviewers and managers.
The safeguards for Fable are very conservative right now since it was banned once. It even bumped me when I was creating something related to my network security.
The hardest part of review tools like this is signal-to-noise — one round of false positives and people start reflex-accepting or muting it. How does Ox decide what's worth flagging on a given PR, and does it dedupe against pre-existing debt so it only surfaces what the change actually introduced?
In my company, we use CodeRabbit with Azure DevOps and it flags a lot of things and mostly they were false positives so eventually we stopped relying on it.