AMA Answer: What pitfalls or mistakes should I avoid, while getting into Context Engineering? What newbie mistake can you help me miss?

Question by Ben Dowen:

What pitfalls or mistakes should I avoid while getting into Context Engineering? What newbie mistake can you help me miss?

Response:

As a newbie, one mistake that I made was trying to put all the context relevant to my work into the AI. I assumed that more context improves results.

I believed that the context windows and tokens were processed correctly until the context window was full. But this is absolutely not the case.

Context rots as the context window increases. AI Models perform poorly when the context exceeds a point (even when it is well within the context window size).  

Drew Breunig nicely outlined several specific ways that longer context can cause performance problems, including:

Only add the context that you need. Be picky in adding context. MCPs also blow up context windows.

In principle, each new additional context source increases the probability of noise, so add them deliberately, not by default. I personally don't use more than one MCP at a time. And, wherever possible, use Agent Skills.

Regularly trim or reset context. Don't let it bloat your system.  More context doesn’t make AI smarter; better context does.