What i keep hearing on calls about claude

Hey there,

David here.

The calls that come in from LinkedIn have taught me more about Claude adoption than our own build did.

Almost everyone I speak to is already using Claude somewhere in their GTM. What I did not expect is how cleanly they sort into three groups, and how different the right next move is for each one.

What separates them is how Claude is operated, and the model barely comes into it. Whether you drive it, or whether it reads from something, or whether it moves before anyone asks.

Each level below is named for what you end up managing, because that is the part that changes.

01Level 1: you manage the prompt

Claude runs in chat. Every session opens with a re-brief, because nothing carries over. Output quality tracks your prompting on the day.

Three tools around it is enough. Prospeo finds the person and verifies the email in one step, Smartlead sends, and a spreadsheet is the CRM. Nothing runs on a schedule, and at this level that is the correct answer. You read every output before it leaves, which is the only guardrail level one needs.

• The tell: you can describe what Claude does for you, and you could not hand it to a colleague and get the same result
• The next step: write down one process you repeat every week. Not the prompt you use. The process

02Level 2: you manage the library

The work moves into a shared repo. Skills become a library Claude reads from, and MCPs connect that library to the live tools. The plan stops being a conversation and becomes a version-controlled file. The ICP work then gets done once and read by everyone, instead of redone by each person.

Three tools becomes eight. HeyReach runs LinkedIn alongside email, Attio holds the relationships, and Supabase holds the data everyone reads from. A spreadsheet stops holding once more than one person writes to it. BounceBan checks the catch-all addresses Prospeo cannot confirm alone.

The first governance question arrives at this level too, and it is smaller than it sounds. Who is allowed to add to the library.

• The tell: someone new can run a process on their first day and produce what the person who built it would have produced
• The next step: put the library somewhere with version history, then make reading it step one of every task

03Level 3: you manage the loop

Claude stops waiting to be asked. Routines grade every live campaign and read every API balance before anyone logs in. The verdict logic sits in code rather than a prompt, so the same numbers always produce the same answer.

The plan changes shape as well. It stops being a file and becomes a system that improves itself. Retros append instead of overwrite, failure modes get checked before launch, and the research compounds across every account.

Eight tools becomes twenty-one at this level, and the loop is what forces it. Someone at level one who hits a dead data provider just tries another one. A routine running at six in the morning cannot, so every layer gets a second source: Apollo and BlitzAPI next to Prospeo, DeBounce behind BounceBan. Signals become their own layer, with RB2B naming the companies already on the site and Apify watching what the market is posting. Sending becomes infrastructure, with EmailBison carrying the volume and iClosed turning a positive reply into a booked call. Claude orchestrates across all of it, with Clay running the enrichment inside outbound and n8n handling the glue between tools. And the ops tools join the stack, because Slack, ClickUp, Notion and Figma are where the work is tracked, and where Claude reaches in.

That last one gets missed the most. The software you filed under back office becomes part of the GTM system, because that is where the work actually lives.

• The tell: something useful happened in your GTM system this morning and nobody triggered it
• The next step: take one judgment you make the same way every time and move it into code

04What decides your level

None of this is set by headcount.

A solo operator can run at level three, because routines and a library do not care how many people read them. A team of twenty can sit at level one for years, with twenty people prompting separately and twenty versions of the same process living in twenty heads.

Headcount changes how expensive level one is to stay on. It does not decide which level you are on.

That is also why the tools are not a shopping list. Three at level one, eight at level two, twenty-one at level three, and every addition is there because the level demanded it. The functions matter more than the logos, and the right combination shifts with budget and stage.

I am curious where you land. Which of the three are you on today, and what is holding the jump.

PS. If you want a Claude system mapped for your setup, book a call with us here