Browse Comments — Clean (de-noised)

Hey Ruben Hassid - last week Claude hallucinated data when I asked it to analyze call recording. My settings specifically told it never to make up information. Apparently, it couldn’t read files in my Google Drive. How can I fix this?

Most people use AI casually. This is how you actually turn it into a working system.

Day 3 is the one most people skip and then wonder why the output sounds generic. Built a voice file for my ghostwriting work and the difference was immediate, not just in tone, but in how much less editing I had to do after. Teaching it what to avoid is half the work. What's the word or phrase you banned first when you set yours up?

Day 2 and Day 3 are the real unlocks here. Once the Project has actual docs and a voice file, the tool starts behaving less like a demo and more like a teammate.

Thank for sharing..claude is going to be game changer now comparison to other ai models.claude is damn good.

Brilliant how you've turned Claude from "fancy chatbot" into actual workflow automation. The connector setup on Day 4 is the real game-changer, most people never get past basic prompting.

The banned words list also trains your own awareness. It teaches you to catch them in your own drafts before Claude even has to filter them out.

Day four connect Gmail. Read last ten emails with contact. Draft answer send ready. No more copy paste.

Day 6 is where most people quietly miss the real ROI. Going from 'I prompt Claude' to 'Claude runs my recurring workflows' is the inflection point — scheduled tasks plus connectors turn it from an assistant you visit into infrastructure that compounds in the background. Solid breakdown, especially the .md voice file on Day 3 — underrated step.

The step-by-step structure here is useful because most people struggle less with the tool itself and more with turning it into part of their actual workflow. What usually makes the difference is not the setup on day one, but whether the workflows become reliable enough to use repeatedly in real work without creating extra review or cleanup later.

on my to-do list

Enterprises achieve scale via deterministic control loops observability telemetry alignment.

This is one of the most practical breakdowns I've seen on getting started with Claude properly. Most people treat it like a fancy Google search - type a question, get an answer, close the tab - and then wonder why they're not getting much value out of it. The real unlock is exactly what you've laid out here: context, continuity, and systems. Day 3 especially resonates with me. Teaching Claude your voice is something most people skip entirely, but it's genuinely the difference between output that sounds like you versus generic AI text. Once you build that voice file, the quality shifts dramatically. The scheduling piece on Day 6 is underrated too. There's something freeing about waking up Monday with your weekly brief already ready, your industry news already summarized. That's when AI stops feeling like a tool you use and starts feeling like infrastructure running in the background. The shift from "using AI" to "building with AI" is real, and this captures it well.

Strong breakdown this makes AI adoption feel practical instead of overwhelming.

I love Claude and need to try Claude design. My only dislike is that when scheduling items to run in Cowork you need to be in Claude for it to run. Has anyone found a way around that? Am I missing something?

This is a solid plan to get up and running fast. Love how practical it is-no fluff, just real steps. I’ve seen how these small wins build momentum. Thanks for sharing this Ruben. I learn so much from your posts.

I love this! It’s bite sized actions that anyone can (and should) do!

but it the same as everyone else is doing. but for sure far better than 99% of the general public. that's the problem.

Thanks for sharing Ruben Hassid !

The Day 1 advice to "Do one real work task" is the most important part. A tutorial shows what is possible. A real task shows what is practical and where you need to provide better context to get a useful result.