> A model doesn’t know that when the sales team talks about revenue they mean commissionable revenue, and the finance team uses something different.
Umm, the model knows the same way humans know: either by looking into slack threads/confluence pages/emails… If your model has access to those things, then it can know. And I foresee that in the near future the model can simply have an incarnation as a Slack member, and when context is not enough, they will simply ask people around (and update docs accordingly)
This underestimate tribal knowledge, wrong assumptions shared across stakeholders, same name ideas with different domains, similarities of words that can mean different things (e.g: revenue) which misleads LLM.
People have a judgment call to say “no this is impossible, when they know something”, but an llm when reads a meeting description of someone that is effusive and truly believes that can be wrong.
As someone that works on data and having seem llm full of context getting wrong assumptions, I’m still looking for the ideal thing.
True, except when it’s not. The correct definition depends on the situation. For example if the if a sales director is pulling numbers to give to a finance director, then the normal sales definition does not apply.
How would the LLM know that?
At the moment I'd say tradition. AI is new enough and scary enough that everyone still feels better with a human in the loop. That being said we rarely find issues when the architecture is planned in advance. Trust is growing.
Im also using the desktop apps (codex/claude). Using the tui doesn’t bring any advantage anymore. I can drag and drop pdfs, images, i have the browser view, can easily switch between conversations, fork them at any point.
I use whatever IDE to review the code, read tests, etc. I used to use the terminal for everything prio AI, but not anymore.
You can drag and drop files in the TUI. You can also paste in images. With tmux you have forking, and tmux windows are fully scriptable, and controllable via your agent. I have them run shell tasks in tmux windows so I can see what they’re doing.
I tried to develop an electron app this way; ubuntu VM (using qemu/multipass) on macos as host with Intellij IDEA. Couldnt make it work properly. Sure, I can install all the toolchain in the VM, and Intellij would recognize it (they have a “remote dev” setup), but for instance for electron apps you cannot just run a headless vm, and there is no convinient way to make a change in your mac, hot reload in the VM, and interact with the electron app in your mac.
The real question for executives is: engineer + AI or PM + AI?
I certainly believe that decent engineers do already have a good degree of knowledge about product topics (it’s like a must nowadays). PM/POs don’t really know enough on the tech side, so they would be clueless even with claude.
But “product” people can’t really build real production software even with claude. Prototypes, sure.
Engineers can (and many do already) deal with product topics. It’s easier for engineers to become more produc oriented than for PM to be able to handle tech topics with AI
I think that's the point. Product only people who didn't build significant engineering understanding aren't going to be replacing anybody and are now at serious risk. Engineering people who did the same and don't want to go find out with customers and the like are also at risk. Smaller teams of people with more cross cutting skills in eng, product, design are going to outperform in this world. everyone else is just vibing their way rapidly to problems imo.
Ai makes it a lot more urgent but pm folks who dropped features from on high never worked as well as pms down in the weeds with engineering in the first place, nor did engineers who didn't want to get out there with customers and understand the why deeply.
Features don’t usually come from product teams. They come from above: some Head Of, VP Of, and the like. Those people only ask, they know very little about execution. They want their wishes come true in a timely manner
Management/Product/Product Engineer - When you are responsible for getting the work done then you look for any and every option to speed it up.
Owner/Platform/Senior Engineer - When you are responsible for ensuring it works and can be maintained then you have to pay attention to how to build it, aka the trade offs
Umm, the model knows the same way humans know: either by looking into slack threads/confluence pages/emails… If your model has access to those things, then it can know. And I foresee that in the near future the model can simply have an incarnation as a Slack member, and when context is not enough, they will simply ask people around (and update docs accordingly)
reply