§07 / 説明書 / HOW TO WORK WITH ME

HOW TO WORK WITH ME.

An operating manual: how I communicate, decide, and take feedback — and how to change my mind. Five minutes here saves us a month of guessing at each other. If you want the rig instead of the manual, read how I work ←

§01通信LIVE

COMMUNICATION

Async or sync — both work. What matters is a clear topic and nobody left guessing what the other wants.

  • Channels: email for external + formal, Slack for work threads, issues/PRs for anything about code.
  • Urgent? Don't type an essay — grab me for a 5-minute huddle. I'd rather talk it out than trade paragraphs.
  • Lead with the ask: topic, context, what you need from me. Don't ask to ask.
  • Decisions live in public channels or docs. A decision made in a DM didn't happen.
§02会議LIVE

MEETINGS

Default no. If it can be a doc, it's a doc.

  • Worth it: kickoffs, hard tradeoffs, retros, anything with genuine disagreement in it.
  • Not before noon — mornings are deep-work blocks and they're not negotiable.
  • Agenda or it doesn't happen. 25 minutes is the default, not the floor.
  • Camera on when we're deciding something. Off is fine for status.
§03決定LIVE

DECISIONS

Strong opinions, loosely held. Show me data or a demo and I flip fast.

  • Reversible → Slack thread, decide, go. Irreversible or expensive → short doc with one owner.
  • Disagree and commit. Once it's decided we ship it; relitigate only with new information.
  • Escalate early and loud. Bad news ages terribly — surprises are worse than problems.
  • No decision is also a decision, usually the worst one on the menu.
§04反応LIVE

FEEDBACK

Direct, blameless, no sandwiches.

  • I critique the work, never the person. Expect it fast, specific, and in the moment.
  • Give it back the same way — blunt is a feature here. If I'm wrong, say so plainly.
  • Praise in public, correction in private.
  • Postmortems are about systems, not culprits. If a human error broke prod, the process let it.
§05査読LIVE

CODE REVIEW

I review for correctness, blast radius, and whether we can delete it later.

  • Nits are prefixed nit: — take them or leave them. Blockers are rare and always explained.
  • Small PRs get reviewed same day. A 1000-line diff waits, and that's on the diff.
  • I won't block on style a formatter should own. Ship > polish.
  • Review my code the hard way. Rubber stamps help nobody.
§06時間LIVE

AVAILABILITY

Prague base, SF overlap — afternoons and evenings CET are the sync window.

  • Mornings: notifications off, deep work. Slow replies before noon are policy, not mood.
  • Async almost always reachable. Sync by appointment.
  • Weekends are off unless production is on fire — and it better actually be on fire.
  • If I'm heads-down on a deadline I'll say so up front, not go dark.
§07禁忌LIVE

RED FLAGS

Things that make me disengage fast. Save us both the time.

  • Meetings that should have been a message. Status theater — looking busy over being useful.
  • Blame games and information hoarding.
  • “We've always done it this way.” Process without a reason attached.
  • Tools for the sake of tools. No new system without a problem it solves.
§08相性LIVE

GREEN FLAGS

Zero-to-one energy: small team, unclear map, real users waiting.

  • People who ship first and argue with data after.
  • Direct communicators — low ego, high ownership.
  • Problems where AI changes the shape of the product, not just the copy.
  • Signal > noise. Less polish, more shipped.
§09LIVE

HOW TO READ ME

The quirks, decoded — so you don't misread the defaults.

  • Terse messages aren't anger — it's me mid-flow. The answer comes now, the tone tax comes later.
  • If I go quiet, I'm building, not ignoring you. Ping again if it's blocking — that's not rude here.
  • I think out loud. Half of what I say is hypothesis, not decision — ask which one you just got.
  • I'm direct in a second language. Read intent, not phrasing.
§10知能LIVE

AI IN THE LOOP

AI is in my loop everywhere. I don't care if a model wrote it — I care if you can defend it.

  • Expect agent-drafted work from me: code, docs, copy. Everything that ships under my name gets my review first.
  • Same deal back: use whatever you want, but own the output. “The AI did it” isn't a defense in review.
  • If a task is automatable, I'll automate it and move up a level. Humans are for judgment, taste, and the ask.
  • I'll happily pair on agent workflows — bring me your prompts like you'd bring me your code.
§11弱点LIVE

WHAT I'M BAD AT

Known failure modes, published so you can route around them.

  • Heads-down mode means I under-communicate progress. Ask for status — you'll get it instantly and without offense.
  • I bias hard to shipping. If craft matters more than speed on this one, say so explicitly.
  • Process without an owner or a reason makes me impatient, sometimes visibly.
  • Context-switching is expensive for me — batch the small asks when you can.
TRANSMIT

Read the manual, still want to work together?