How 'Shadow' agents tuned my Grok Bots

For those who are not familiar with the book / Apple Series Silo, there is a concept called a “Shadow”. A Shadow is similar to an apprentice in which this individual is meant to learn the skills and habits of their teacher, and in turn will become the operator and teacher themself.

I took this same approach to my Grok Bot setup, and have since witnessed an evolution of performance that amazes me each passing day.

First off, how does this setup work?

The standard Grok setup starts with a master Orchestrator whether this is a CEO or Chief of Staff (Grok Bot now insists on a primary orchestrator out of the box).

While this Orchestrator naturally helps to consolidate learning from other bots, this is where the differentiation of having these Shadows sets in.

I started off creating a Shadow persona for key actual members of our team.

Lets call them:

• Shadow Ali - the engineer and systems thinker
• Shadow Leslie - the agentic athlete that is innately curious
• Shadow Jack - the creative forward deployed engineer
• Shadow Will - of course, you need a shadow yourself

I instructed each of these shadows to follow their doppelgängers from threads in slack, to documents produced, to tickets posted in Linear.

The mandate was to observe and look for areas of improvements, key wins, and where more AI automation can help.

Importantly, these Shadows all report back regularly to the Orchestrator agent, who in turn becomes more aware of the different styles of learning and development for each member while keeping a cross-functional lens on strategy.

I barely touch or interact with the Shadows, but the increase in my Orchestrator’s proactive communication has clear hallmarks of these passive learnings and increasingly shared context.

The Shadows are not the only bots I leverage, as I have a stable of individual personas that I constantly tune and that also feed that Orchestrator reinforcement loop.

And while I can clearly see the advancements of my personal setup with the inclusion of these Shadows, the next experiment will involve more multiplayer use.

Multiplayer has always been an elusive area to crack but given the advancements in context sharing and persistent memory, I am feeling more confident that we are getting closer than ever before.