Iman Gadzhi opens with a confession that will sting for anyone paying a monthly AI subscription: most people are using less than 2% of what Claude can actually do. The uncomfortable reality is that opening Claude, typing one question, and closing the tab is just an expensive version of Google. For anyone who has watched credits evaporate without much to show for it, that framing lands hard. After spending more than a thousand hours testing Claude across his own companies, Gadzhi mapped everything into four tiers: beginner, novice, intermediate, and pro, and the gap between where most people start and where the tool can actually take them is significant.
The rule most people skip entirely
The foundation of everything Gadzhi teaches is what he calls the 10-80-10 rule. The user brings the first 10%, which is the context Claude needs to do the job properly. Claude handles 80%, the actual work. The user returns for the final 10% to check and tighten the output. The rule sounds simple, but most people violate it at step one by giving Claude almost no context at all. Asking for ‘five marketing ideas for my business’ returns five ideas that could apply to any business on earth. Adding a spending cap, a description of the actual product, and a constraint like running a side hustle alongside a full-time job gets something built around a real situation. Two extra minutes on the prompt, Gadzhi argues, can make the result ten times more useful.
Beyond the prompt itself, the beginner tier covers two more habits: uploading documents, screenshots, or even photos of handwritten notebooks instead of trying to describe their contents in text, and telling Claude to ask clarifying questions before it starts rather than letting it fill gaps on its own. Each of these reduces the back-and-forth from eight or nine iterations to two or three.
Where Claude stops forgetting you and starts doing the work
The novice tier addresses the single most frustrating limitation: every new chat starts from zero. The fix is memory, found under settings, where two toggles let Claude search previous conversations and generate a persistent profile as you work. Gadzhi also points to an import function that carries context over from other AI tools, which he describes as solving the ‘climbing a mountain just to get back to where you were’ problem for anyone switching from a platform they had been using for months.
The model-matching insight at this tier is where real money gets saved. Gadzhi’s team burned through expensive credits before figuring out that running the most powerful model for every task is the equivalent of handing a C-suite executive a five-minute intern job. Claude’s model lineup at the time of writing runs from Haiku for quick back-and-forth, through Sonnet as a daily driver, up to Opus for multi-step analysis, and then Fable at the top. Gadzhi is direct about Fable: ‘this thing is good, like scary good, but it costs a lot of credits to use, so early on I would actually leave it alone.’
The intermediate tier moves Claude off the island. Connectors in settings link Gmail, Slack, Google Drive, Notion, and other apps so Claude can pull together what happened while someone was away, draft replies to the emails that actually matter, or find open blocks in a calendar before the morning coffee is made. A Chrome extension lets it read open browser tabs without any copying, which he illustrates with a six-contractor comparison: instead of hours reading through six proposals with buried fine print, Claude surfaces the one worth calling back in about five minutes. Co-work extends access further, letting Claude read, edit, and create files inside designated folders on the machine itself.
The pro tier is where Claude stops working on things that already exist and starts building things that do not. Using Claude Design, a user describes a client-tracking dashboard. Design produces a clickable prototype to review and adjust before anything is built. That plan exports directly to Claude Code, which writes and tests the underlying software. Recurring tasks are then handled through skills, saved workflows that run the same way every time, and scheduled tasks that fire automatically. Gadzhi’s summary of the endpoint is precise: ‘you described a tool that didn’t exist, Claude designed it, it built it, and now it is keeping the whole thing running without you even in the room.’
The Remarkable notebook sitting in a chat window
Gadzhi mentions uploading his Remarkable digital notebook into Claude after long writing sessions on planes, sometimes hours of freeform thinking captured in one file, handed over as raw context for whatever he is working on next.
The four tiers build on each other in a single direction: less explaining, less manual carrying of information between windows, and eventually a small ecosystem of skills where the output of one feeds directly into the next. The 10-80-10 rule that starts the whole thing is still running at the top of the pro tier, just automated.
Wake up to Only Happy News!
