CLAUDE CODE MEETING PHRASES — CHEAT SHEET

Quick-grab lines for presenting how I use Claude Code. Scroll to section, pick a line, say it.

1. Self-Positioning: “Heavy User, Not Power User”

• “I use Claude Code for basically everything at this point. Not in a super sophisticated way — just high volume, all day, every task.”
• “I’m probably one of the heaviest users on the team, but I wouldn’t call myself a power user. I just throw everything at it.”
• “My setup isn’t fancy. I just use it constantly. Like, it’s open all the time, it’s part of every workflow I have.”
• “I’m not doing anything crazy with custom toolchains or anything. I’m just… using it for everything. Quantity over cleverness.”
• “If there’s a spectrum from ’tried it once’ to ‘built a framework around it’ — I’m somewhere in the middle, but I use it more than anyone I know.”
• “I wouldn’t say I’m the most technically advanced user. But in terms of hours spent working with it? Yeah, it’s a lot.”
• “Think of it like — I’m not a chef, but I cook every single meal at home. That’s my relationship with Claude Code.”

2. Experience Positioning: “Least Work Experience, Not Least AI Experience”

• “I know I’m the most junior person here in terms of career. But I’ve been using AI tools since way before this job — like, paying out of pocket for them.”
• “I’ve been on this ride since GPT-3. Paid my own money for API access, Claude Pro, all of it. This isn’t new to me.”
• “I might have the least work experience on the team, but I definitely don’t have the least AI experience. I’ve been deep in this for years.”
• “I was spending my own money on AI subscriptions before most people had heard of ChatGPT. So the tooling part — that’s not where I’m junior.”
• “Don’t let my tenure fool you — I’ve been through the whole evolution. GPT-3, ChatGPT, Claude, Copilot, Claude Code. I’ve used all of them seriously.”
• “Where I’m junior is the enterprise engineering stuff. Where I’m not junior is figuring out how to get AI to actually do useful work.”
• “I’ve literally been an AI tool power user longer than I’ve been a professional software engineer. That’s just the timeline.”

3. The 3-Level Framework (from Proshat)

• “Proshat has this framework I really like. Level 1 is interactive — you’re going back and forth with the agent. Level 2 is a closed loop — you set it up, it works, you review. Level 3 is fully autonomous, shippable output. I’m aiming for Level 2.”
• “The way I think about it — Level 1 is pair programming. Level 2 is delegation. Level 3 is having a junior engineer you trust completely. I’m solidly targeting Level 2.”
• “There are three levels. One is chatting with the agent. Two is — you give it the task, it does the work, you review the output. Three is it ships without you. I’m working toward a solid Level 2.”
• “Level 1: you’re driving. Level 2: it’s driving, you’re reviewing. Level 3: it’s driving and you’re not even in the car. I want to be a reliable Level 2.”
• “The framework goes: interactive, closed-loop, autonomous. I’m at the point where I set things up, let it run, and then review. That’s Level 2.”
• “I don’t want to sit there prompting back and forth all day — that’s Level 1. I want to give it context, let it cook, and check the result. That’s the Level 2 I’m going for.”

4. Input Sources

• “My inputs are basically everything I’m already looking at. Slack threads, DMs, Linear tickets, GitHub PRs, Canvas docs — all of that goes into the agent as context.”
• “I feed it whatever I’m working from. If it’s a Slack conversation, that goes in. A Linear ticket? That goes in. PR comments? Yep.”
• “The input side is pretty simple — it’s all the stuff I’d normally be reading anyway. Slack, Linear, GitHub, docs. I just pipe it into the agent.”
• “Anything that would normally live in my browser tabs becomes input. Slack threads, Canvas docs, GitHub issues, my own local notes.”
• “I also keep personal local notes and task management stuff that I feed in. It’s not just team artifacts — it’s my own scratchpad too.”
• “The beauty is I don’t have to reformat anything. Slack thread? Copy it in. Linear ticket? Paste the link. PR diff? Just give it the context.”
• “Think of it as — whatever I’m reading to understand a task, the agent is reading the same stuff.”

5. “I Don’t Care About the Processing”

• “I don’t micromanage the middle. I set up the context, I set up what I want, and then I let it work. I review the output.”
• “The whole point for me is — I don’t care how it gets there. I care about the input and the output. The processing is the agent’s problem.”
• “I’m not sitting there watching it think. I give it the task, I go do something else, I come back and check what it produced.”
• “It’s like — I don’t watch my code compile either. Same energy. Set it up, let it run, check the result.”
• “The middle part is a black box to me and I’m fine with that. I’m optimizing for my time, not for understanding every token it generates.”
• “My job is to give it good input and catch bad output. Everything in between? That’s why I have an agent.”

6. Output: Slack Replies & Drafts — Must Sound Like ME

• “When it drafts a Slack message, it goes out under my name. So it can’t sound like AI. No ‘certainly!’ No ‘I’d be happy to help.’ That’s an instant tell.”
• “I’ve spent a lot of time training the tone. The messages have to sound like me — casual, direct, the way I actually talk on Slack.”
• “If someone reads my Slack message and thinks ‘did AI write that?’ — I’ve failed. The whole point is it sounds natural. Like I typed it.”
• “The bar is: would my teammates notice? If yes, it’s not good enough. I’ve iterated on this a lot.”
• “I’m not just generating text — I’m generating text in my voice. Bay Area casual, a little blunt, no fluff. That’s what ships.”
• “The worst thing an AI message can do is sound polished. Real Slack messages are messy, short, sometimes half a sentence. That’s what I want.”
• “I basically taught it to write like a mid-20s engineer on Slack, not like a customer service bot. That took work but it’s worth it.”

7. Output: PRs and PR Review Comments

• “For PRs, the agent generates the code and the description, but I review everything before it goes up. That’s a natural checkpoint.”
• “I’m the QA layer. The agent writes the PR, I read through it, make sure it’s correct, and then I submit. Nothing goes out unreviewed.”
• “PR review comments are the same — it drafts them, I check them, then I post. I’m not just rubber-stamping.”
• “The nice thing about PRs is there’s already a review step built into the workflow. The agent does the heavy lifting, I do the quality check.”
• “It generates, I validate. For PRs especially, that loop is really clean — write, review, submit. I’m always in the loop before it hits GitHub.”
• “I think of PRs as the safest output type because there’s already an approval process. Even if something slips past me, there’s still code review.”

8. Output: Learning Notes / Study Material / Shareable Docs

• “One output I really value is docs. Study material, reference guides, things I can share with the team. The agent is great at producing those.”
• “I use it to capture knowledge — like, if I just figured something out, I’ll have it write up a doc so other people can benefit too.”
• “A lot of what comes out isn’t code — it’s documentation. How-to guides, architecture notes, troubleshooting references.”
• “The shareable docs are honestly one of the highest-value outputs. I can learn something once and turn it into a resource for the whole team.”
• “It’s really good at taking a messy investigation and turning it into a clean, readable doc. I do that a lot.”
• “I basically use it as a knowledge capture engine. Every debugging session, every investigation — there’s usually a doc at the end.”

9. The Meta-Loop: Outputs Become Inputs

• “Here’s the thing that makes it compound — the docs I produce in one session become the context for the next session. Outputs become inputs.”
• “It’s a feedback loop. I generate a doc, next time I start a session, that doc is part of the context. So it gets smarter about my work over time.”
• “The meta-loop is what makes this really powerful. It’s not just one-shot — the outputs feed back in and the quality keeps going up.”
• “I’ve built up this library of context docs that I feed into every session. Each one started as an output from a previous session.”
• “This actually works better than Claude Code’s built-in session handoff. My own docs, my own context, carried forward manually. More reliable.”
• “Think of it like — each session leaves behind artifacts, and those artifacts make the next session better. It compounds.”
• “The first session is always the hardest. By the fifth session on the same topic, the agent basically knows what I know.”

10. Wrap-Up / Summary Lines

• “So basically — heavy usage, not fancy usage. I feed it everything I’m already reading, I don’t micromanage the middle, and the outputs have to sound like me.”
• “TL;DR: I use Claude Code as a closed-loop assistant. Context in, reviewed output out, and the outputs feed back into future sessions.”
• “The short version is: I’m aiming for Level 2. Set it up, let it work, review the result. And everything compounds over time.”
• “If I had to summarize in one sentence: I use AI to handle the throughput so I can focus on the judgment.”
• “Basically — I’ve been doing this long enough to know what works. High-context input, hands-off processing, human-sounding output. That’s the loop.”
• “The key takeaway is: it’s not about being clever with prompts. It’s about feeding it good context and having high standards for what comes out.”
• “End of the day — I’m a heavy user because it genuinely makes me faster. Not in theory. In practice, every day.”

RECOVERY PHRASES

“Can you repeat that?”

• “Sorry, can you say that again?”
• “Wait, say that one more time?”
• “I missed that — what was the last part?”
• “Can you repeat that? I want to make sure I got it right.”
• “Hold on, I didn’t catch that — one more time?”

“Let me think”

• “Hmm, let me think about that for a sec.”
• “Good question — give me a moment.”
• “Yeah… let me think about how to phrase this.”
• “That’s a good point, let me sit with that for a second.”
• “Hold on, I want to give a good answer here.”

“What do you mean by…?”

• “When you say [X], what exactly do you mean?”
• “Can you clarify what you mean by that?”
• “I want to make sure I’m understanding — are you asking about [X] or [Y]?”
• “What do you mean by [X] in this context?”
• “Sorry, can you unpack that a little? I’m not sure I follow.”

TRANSITION PHRASES

“And the cool part is…”

• “And the cool part is —”
• “But here’s where it gets interesting —”
• “And what’s really nice about this is —”
• “The part I think is actually cool —”
• “And this is the part that surprised me —”

“But the thing I care most about…”

• “But honestly, the thing I care most about is —”
• “At the end of the day, what matters to me is —”
• “The part I’m most focused on is —”
• “But the thing that actually matters here is —”
• “What I really optimize for is —”

“Oh and one more thing…”

• “Oh, and one more thing —”
• “Actually, one thing I didn’t mention yet —”
• “Oh wait, I should also say —”
• “One more thing that’s worth mentioning —”
• “And this is kind of a side note but —”

FUN FACT / PERSONAL INVESTMENT

🎤 The Money & Time Investment

• “Fun fact — I’ve probably spent more personal money on AI tools than anyone on this team. Like, my own credit card, not the company’s.”
• “I’ve been doing this since before GPT-4 was a thing. Paid out of pocket for every AI tool along the way. It’s basically a hobby that became my job.”
• “I might have the least work experience here, but I’ve been messing with AI coding tools since they were terrible. Like, Copilot-before-it-was-good era. Paid for all of it myself.”
• “One thing about me — I’ve been on this AI journey a long time. Through every generation. And yeah, a lot of personal money went into that.”
• “I’ve been an early adopter for basically every AI coding tool that’s come out. My credit card has the receipts.”

COPILOT DEVEX TEAM / CAM / INTERNAL TALKS

🤝 Introducing the Connection

• “Oh also — the Copilot DevEx team has done a ton of work on AI coding workflows. I’ve been talking to Cam from their team whenever I hit something interesting.”
• “So there’s a team internally — Copilot DevEx — that’s been going deep on this. I talk to Cam from that team pretty regularly, we swap notes on what’s working.”
• “One resource I’d flag — the Copilot DevEx team. They’ve put a lot of thought into AI coding. I’ve been bouncing ideas off Cam over there.”
• “If you want to go deeper on this — I’ve been chatting with Cam from the Copilot DevEx team. They’re way ahead on the tooling side.”
• “I don’t want to take all the credit here — a lot of what I’ve learned comes from talking to folks like Cam on the Copilot DevEx team.”

📢 Pitching the Internal Talks

• “And actually — one thing if you’re interested — they host a regular talk where people show their AI workflows internally. I can share the meeting invite link.”
• “Oh and they run a recurring session where people demo their AI coding setups. If that sounds interesting I can forward the invite.”
• “They’ve got this regular internal meetup — people show how they actually use AI in their day-to-day. I can drop the invite link in the channel if anyone wants it.”
• “If you want to see what other teams are doing — Copilot DevEx runs a show-and-tell for AI workflows. I’ll share the link after this.”
• “One thing I’d recommend — there’s an internal series where people present their AI setups. Low-key, practical. I can send the invite if you’re curious.”