You leave with a vibe. Not decisions. Not owners. Not dates. Fix it with a 10-minute SOP you run after every call. Then you can bring AI in and it actually helps. Here’s the simple workflow: 1) Use one shared doc every time: “Decision Log + Action List” Two sections. Same order. No creativity. Decision Log (copy/paste this header) - Decision: - Why: - Trade-off accepted: - Who owns the decision: - When we revisit: Action List (every line must have all 4) - Owner: - Due date: - Definition of done: - Next check-in date: 2) Do a 5-minute “capture pass” immediately after the call No polishing. Just dump the raw notes, screenshots, timestamps, whatever. 3) Then use AI for one job: convert raw notes into your template Give Claude your meeting transcript. The instruction is simple: “Fill the Decision Log and Action List. If any field is missing, write ‘MISSING’ and ask one clear question.” Now you get something you can actually review. 4) Do a 2-minute “MISSING sweep” If “Owner” is missing, the task doesn’t exist. If “Definition of done” is vague, it becomes a ghost task. If there’s no “Next check-in date,” it won’t survive the week. The punchline: your team doesn’t need more meetings. You need a repeatable way to turn talk into commitments. Every. Single. Time.

If you can’t say the visual’s job in 10 words, you’ll keep shipping pretty nonsense. It’s because you’re asking for a picture, when you actually need a repeatable visual system. Do this the next time you need a graphic for a post, a deck slide, or a thumbnail (Gemini works great for this). Step 1: Decide the job of the visual in one sentence. Example: “Make this post skimmable for busy operators.” If you can’t say the job, you’ll keep generating pretty nonsense. Step 2: Pick one layout you can reuse for 30 days. Stop reinventing design every time. Pick one: - Big headline + simple icon - Numbered checklist card - 2-column “Problem / Fix” - Quote card with a strong border Step 3: Give Gemini a “visual spec” instead of vibes. Copy/paste this and fill it in: Asset: (LinkedIn 4:5 image, 1080x1350) Topic: (what this is about) Audience: (who it’s for) Message: (the one takeaway) On-image text: (max 10 words) Style words: (pick 3: clean, bold, calm, technical, playful, editorial) Colors: (2-3 hex codes) Fonts: (any preference, or “clean sans-serif”) Composition: (centered, lots of whitespace, left-aligned text, etc.) Brand element: (one repeated thing: thin border, corner tag, small icon style) Avoid: (no faces, no clutter, no gradients, no fake “3D”) Step 4: Generate a set, not a single image. Ask for 8 variations of the same spec: - 4 with icon-led layout - 4 with text-led layout You’re trying to find a “house style,” not win the lottery. Step 5: Lock the style with one keeper. When you get one that’s close, tell Gemini: “Use this exact style for 5 more images with different headlines. Keep the same layout, colors, and spacing.” Now you’ve got a system. Here’s a real example you can steal for your next ops post: Asset: LinkedIn 4:5 image, 1080x1350 Topic: Weekly team update Audience: founders + operators Message: Updates should reduce questions, not create them On-image text: “A weekly update that stops Slack chaos” Style words: clean, calm, structured

https://www.youtube.com/watch?v=uSTGNHGFOAo

Your support inbox doesn’t feel “busy”. Randomness is screwing up your productivity. A password reset sits next to a cancellation threat. A serious bug report gets buried under “where’s my invoice?” Claude Routines is quite popular so just use that. Set it to trigger every single day in the morning and make it sift through your emails and... Here’s the version that works in real life: 1) Choose 5 buckets you’ll actually use Keep it boring so your team sticks to it. - Billing / invoice - Access / login - Bug / broken - Cancellation / refund risk - Pre-sales / pricing 2) Write the priority rules like you’re explaining them to a new hire If you skip this, the sorting will feel “off” and everyone will ignore it. Examples you can steal: - Mentions “cancel”, “refund”, “chargeback”, “fraud” → high - Pre-sales + team size, budget, timeline, “need this by…” → high - Bug + “blocked”, “production”, “can’t ship” → high - “how do I…”, “where is…”, “can you resend…” → normal 3) Make Claude return the same fields every time Consistency is the whole point. You want routing and reporting without extra work. Have Claude output: - Bucket - Priority (high / normal / low) - One-sentence summary in plain words - What the person is asking for (refund, ETA, workaround, discount, etc.) - Suggested next step for a human agent 4) Route based on bucket + priority This is where the chaos starts turning into a queue you can trust. - High → ping + assign now - Normal → right queue - Low → wait, or send a quick “received” reply 5) Give agents a first-reply frame per bucket A frame beats a blank page. Refund risk: “Got it. Tell me what happened and what you expected. If we messed up, I’ll fix it today.” Bug/blocker: “Quick triage: what were you trying to do, what happened, and what’s the impact right now? Paste the error text if you can.” Treat every message like it has the same urgency and your day becomes a slot machine. Make the inbox predictable. Let Claude do the sorting. Keep humans for the decisions.

If AI feels “mid” to you, it’s probably because you’re using it like a vending machine. You type: “Write me a landing page” or “Make me a content plan.” It spits out beige. Then you blame the model. The fix isn’t a smarter prompt. You need better context. Here’s a simple way to feed context that actually changes the output in one go. Copy this and fill it in. Keep it saved as your “Context Card” and paste it before any real request: 1) What you’re making (one sentence) Example: “A landing page for a 2-week onboarding sprint for busy founders.” 2) Who it’s for (pick one person, not a market) Example: “A founder who has 12 tabs open, 40 Slack pings, and keeps pushing onboarding ‘to next week’.” 3) The moment they’re in (what happened right before they found you) Example: “They just hired their first ops person and realized nothing is documented.” 4) What they already tried (and why it didn’t work) Example: “They tried a Notion template dump. It turned into a graveyard.” 5) Your taste rules (3–7 bullets max) Example: - Short sentences. No hype. - Practical, not philosophical. - Assume they’re smart and tired. - Give examples, not frameworks. 6) Proof you can actually claim (real, plain) Example: “Includes a 30-minute kickoff, 5 docs, and a weekly review checklist.” 7) The one action you want them to take Example: “Book a call” or “Start the sprint” or “Reply with ‘ONBOARD’.” Then ask for the thing you want: “Using the Context Card above, write a landing page with: headline, subhead, 3 sections, FAQ, and a final CTA. Write v1. Leave placeholders where you need specifics.” Two important notes: - “Taste rules” is where your voice lives. Most people skip it. That’s why everything sounds the same. - Let AI write v1 on purpose. Your job is v2–v10: cut half, add one real example, remove anything you wouldn’t say out loud. If your outputs keep sounding generic, don’t prompt harder. Describe the person, the moment, and your taste. That’s the whole difference.