I have a confession. For way too long, my Claude workflow looked like this: open a tab, type a question, copy the answer, close the tab. Repeat forty times a day. It felt productive, but honestly it was just a fancier search box.
So when I spotted this post from a LinkedIn creator who’s been building with Claude across an actual team, it hit a nerve. The original poster’s argument is simple: most people are getting about 10% of the value from Claude because they treat it like a slightly smarter Google. No context. No continuity. No real output. And then they wonder why the productivity gains everyone talks about never show up.
What made me trust this one is that the author admits they got it wrong too. Half their team was using Claude regularly and still not seeing the expected results. They were prompting well. They just weren’t working with the tool, they were querying it. Once they started building systems around it, output per hour doubled, then tripled.
The post lays out seven steps for making that shift. I’ve unpacked each one below, with the reasoning behind it, so you can start applying this today.
🎯 The 7-step system for getting real work out of Claude
- Choose the right model first. The expert’s rule of thumb: Sonnet for fast daily work, Opus for complex reasoning, Claude Code if you’re building software, and Research mode when you need deep analysis. Their line for using the wrong one is perfect: it’s like hiring a surgeon to do admin. Why it matters: picking a heavyweight model for a quick email rewrite wastes time, and picking a fast model for a tricky strategy doc gets you shallow output. Match the tool to the job before you type a single word.
- Master your prompt structure. The author uses a fixed order every time: Role, Goal, Context, Constraints, Output format, Success criteria. In that order. Why it matters: Claude performs like a very capable new hire. Hand it a vague request and you’ll get a vague result. Hand it a proper brief and the quality jumps. The creator puts it plainly: the better the brief, the better the output. Every time. Try it on your next task and compare.
- Create Projects. This is the step the post’s author says most people leave serious value behind on. Projects let Claude hold long-term context across sessions. No re-explaining. No repeating yourself. Your clients, your workflows, your voice, all retained. Why it matters: every fresh chat without a Project means you’re rebuilding context from zero. That’s the hidden tax that kills productivity. Set up one Project per client or major workstream and the tax disappears.
- Build Artifacts. Claude isn’t just for answers, it’s for deliverables. The expert lists documents, dashboards, SOPs, apps, checklists, and websites. Actually usable. Actually shareable. Why it matters: when you stop at “give me an answer” you still have to go make the thing. When you ask for the artifact directly, the finished piece lands in front of you, ready to hand off.
- Connect Memory. Save your brand voice, your goals, your preferences, and your work context. The author’s point: the more Claude knows about how you operate, the less prompting you do over time. Why it matters: this compounds. Week one you’re still explaining your tone. Month three, Claude already knows it, and your prompts shrink to a single line.
- Build Custom Skills. Think of these as reusable expertise packs for content, marketing, business, or engineering tasks. Stop doing the same setup work every session. As the creator puts it: build it once, run it forever. Why it matters: if you catch yourself pasting the same instructions into Claude for the third time this week, that’s a Skill waiting to be built.
- Connect your tools. Google Workspace, Notion, Slack, GitHub, CRMs, internal knowledge bases. Real-time context, fewer manual tasks, better workflows. Why it matters: copying data into Claude by hand is the last bottleneck. Once it can read your calendar, your docs, and your codebase directly, it stops being a chat window and becomes a teammate.
💡 Why this order matters
What I appreciate about this innovator’s framework is that the steps stack. Model choice and prompt structure fix the quality of a single conversation. Projects and Memory fix continuity between conversations. Skills and tool connections fix the repetitive grunt work around conversations. Skip the early steps and the later ones don’t have a solid base to sit on.
🛠️ How to start this week
- Day 1: Write one prompt using the six-part brief structure. Notice the difference.
- Day 2: Create a Project for your most active client or initiative and drop in the background docs.
- Day 3: Ask for an artifact instead of an answer. A checklist, an SOP, a one-page doc.
- Day 4: Add your voice guidelines and preferences to Memory.
- Day 5: Spot one repeated task and turn it into a Skill.
The people building real leverage with AI right now aren’t the ones prompting harder. They’re the ones building systems.
That closing line from the post’s author is the whole idea in one sentence. Prompting harder gets you marginal gains. Building the system around the tool is where the doubling and tripling happens. I’ve been rebuilding my own setup around these seven steps and the difference is already obvious!
The full post has a bit more color on the team story behind all this. Definitely worth a read on LinkedIn.