Claude Code · as a personal assistant

I Gave My AI
a Brain
and Hands.

I turned a coding tool into an assistant that knows my world and actually does the work — not just answers.

poomiphat.commy blog · about me
scroll↓
01
The gap

How many hours a week do you lose — re-explaining context, and doing the grunt work by hand?

An LLM is a brilliant brain in a jar. It answers — but it doesn't know your world, and it can't touch your tools.

Plain chat — no context
Re-explain your role, business, goals — every single session.
→ wasted effort · generic answers
→
Once it knows your context
It already has your world. Skip the setup, every time.
→ sharper answers, instantly
02
The whole idea

A real assistant = brain + hands + an LLM in the middle

Brain
Context
knows your world
+
Hands
Connected tools
acts on it
→
LLM core
Decides & does
real output

Everything after this is just filling in the two halves.

03
Half one · the brain

Everything it knows about my world

Context — structured like a filesystem, fed into the brain.

~/brain/context/tap a file →
business-knowledge.md
how the business actually works
Teams & responsibilitiesScope of workRevenue modelKey metricsProduct roadmapOrg structure
meetings/
what was said, decided, owed
Ops meeting asks since AprDecisions & ownersBlockers raisedAction itemsWeekly syncs
articles/
things worth keeping
Saved researchFrameworksIndustry referencesHow-to guides
stakeholders/
who's who, who cares about what
Who's whoWhat each cares aboutReporting linesRecent updates
LLM core
04
Half two · the hands

Tools it can actually operate

Connected through MCP — it reads in and writes out, both ways.

Confluence
Jira
Miro
the brain
Lovable
Database
Airtable
▸ real run "Where are our Lovable credits burning?"
1 pulled prompt patterns from 30 Lovable projects→ 2 analyzed how everyone prompts→ 3 wrote a credit-saving tips guide→ 4 + a specific fix pattern per project→ 5 built the share-out deck
05
Teach once · reuse forever

From a messy idea to dev-ready — taught once

Grill PM

/grill-pm

Interrogates a rough idea into a clear functional spec.

To tickets

/to-tickets

Slices that spec into dev-ready tickets.

Grounded in your repo — tickets match how your code actually works. github.com/psisquare/skills ↗
06
Your turn

Your job is also
context + tools

  • Brain = what you know that the AI doesn't yet.
  • Hands = the tools you already live in.
  • Pick one repeatable task → teach it once.
Lessons learned
01
It's all context management
The more your AI knows about you, the better the results — and the bigger the productivity gain.
02
No perfect system on day 1
It gets better when your hands get dirty — build it, kill what doesn't work, iterate.
poomiphat.com/blogread the full write-up — Personal OS, ep.1
Personal OS · a brain + a pair of hands for your work · poomiphat.com