Opening Note

The hive

Two years ago I had a thoughtful, very organized system of external tools and personal behaviors. Today I have an owned, central hive of integrated tools — some I built, some I plugged in, all of them speaking the same language. I didn't have the blueprint when I started. I iterated my way into it. Here's how.

The Timeline

Two years, six phases — each one made possible only by the one before it.

24 Months Ago
Claude Chat  ·  ChatGPT

AI as research surface — question and answer, ephemeral.

ReplacedGoogle searches and long reading sessions.

LearnedConversations evaporate when you close the tab. The value isn't in the answers — it's in being able to come back to the thinking.

18 Months Ago
+ Claude Chat Projects  ·  ChatGPT Projects  ·  Perplexity Research

AI as idea container — ideas got a home.

ReplacedScattered chat tabs and lost threads.

LearnedWithout knowing where AI was going, I was creating unique assets. Those assets only had value because they had a home.

12 Months Ago
+ Claude Chat Projects as source synthesizer

AI starts reasoning over MY sources, not just public ones.

ReplacedManual cross-referencing between my own files.

LearnedThis is when assets started compounding instead of repeating. Old work made new work more valuable, not less.

4 Months Ago
Claude Chat  ·  Claude Cowork  ·  Perplexity Research

AI as tool replacer — retire the legacy stack.

ReplacedExcel  ·  Google Drive functions  ·  parts of Notion.

LearnedMost people layer AI on top of their existing tools. The bigger move is deciding which legacy tools can die.

2 Months Ago
+ Claude Cowork (Dispatch  ·  Scheduled Tasks  ·  Code  ·  Projects  ·  Skills)  ·  Gemini

AI as workflow operator — autonomous-with-checks, not interactive prompting.

ReplacedMe typing prompts. Me clicking through tools.

LearnedThis phase wasn't possible six months ago for non-coders. Cowork is the unlock. The shift from "AI helps me work" to "AI does work while I sleep" is technological AND operational.

1 Month Ago
Same stack, now optimized — off-peak compute  ·  library integrations  ·  MCP bridges

AI as critical infrastructure — backups, security, scheduling, failure modes.

ReplacedTreating AI like a productivity helper.

LearnedWhen AI moves from "I use it" to "my work runs on it," you protect it like infrastructure. Compute is a budget. Monitoring is mandatory. Failure modes are real.

Three Takeaways

What the hive taught me along the way.

The human doesn't go away — the work moves.

Capture and enrichment friction drops to near zero. Editorial judgment, voice, taste — those stay mine. I do less typing and more deciding.

Set-it-and-forget-it is a lie.

I'm in the workflow every day. Last week the save flow on my capture tool broke silently for three days. I caught it because I was using it, not because the system told me. The automation works because someone is paying attention to where it breaks.

Compute costs real money.

Off-peak scheduling is a discipline. Treat your AI infrastructure like any other budget. Allocate it on purpose.

Don't predict which model wins.

Own your data, your taxonomy, your workflow patterns.

The rest plugs in.