Let’s say I spend half a day with an agent debugging some weird infrastructure problem. Eventually we figure it out.
Three months later someone else in the company hits almost the same problem.
Today there’s a decent chance they’ll never know I already went through it, because the useful context is sitting inside my conversation with the agent.
Are people seeing this happen?
I’m curious how teams using agents a lot are dealing with it. Do you push useful stuff into docs/RAG automatically? Have the agent decide what’s worth keeping? Just rely on people to document things?
Automatically sharing everything sounds terrible too — you’d end up with huge amounts of junk.
Would love to hear how people are approaching this in real teams.
Let’s say I spend half a day with an agent debugging some weird infrastructure problem.
Let’s say you did that with GPT x — for example, GPT-5.
When GPT x + 1 (GPT-6) comes along, perhaps that same problem takes 10 minutes instead of half a day.
By the time GPT x + 2 (GPT-7) arrives, the process may be automated enough that you don’t really encounter the underlying problem yourself anymore.
So I wouldn’t necessarily expect people to report seeing this happen. The capability of the models is changing at the same time as the problems and workflows they are being used to solve.
What looks like a persistent problem from one model generation can effectively disappear with the next few generations—not necessarily because the underlying infrastructure changed, but because the model became capable enough to work around or solve it.