I’ve been building autonomous agents for weeks and context rot is a major problem I keep coming back to. Here’s some data and what’s been working for me.
“1 million context?” eh? The longer your agent runs, the dumber it gets

Hi! I’m Erik MacKinnon — Head of Revenue at Wynter, career entrepreneur, and an AI-native GTM operator building agentic systems that make B2B revenue engines go faster.
These days I’m in the skunkworks designing AI-native GTM systems – agentic workflows, autonomous agents, LLM-powered revenue operations – and writing about what actually works when you put this stuff into production.

I’ve been building autonomous agents for weeks and context rot is a major problem I keep coming back to. Here’s some data and what’s been working for me.

I keep watching agents nail steps 1-4, then silently drop 5 and 6 while completing 7. These probabilistic failures are common, and a pain in the ass. Here’s what I’m doing to try to fix it.

It’s a simple phrase, isn’t it? “Now, not how.” Yet these three words can be something special in the right hands.

These are my notes for “The JOLT Effect” by Matt Dixon and Ted McKenna. This isn’t a review, it’s my notes. So… it’s long. Enjoy!

Do you work on a team? In my experience, in B2B, many think so. They’re mistaken. It’s a costly problem in revenue organizations.