Most Claude Code “experts” have never carried a bag
If someone’s preaching a tool they’ve never had to hit quota with, you’re watching theater.
Watched a stretch of LinkedIn this week where people sell Claude Code workflows “for sales.” Most of them have never carried a bag or shipped production code. They’ve got LLM-grabbed theory dressed up as field experience. Here are my 2 cents about why I’m skeptical, and the one thing that actually moves the needle.
🎯 THE ONE THING
The loudest voices on AI tooling for sales are usually neither salespeople nor deep technical people…
They’ve spent a week with Claude Code, watched it spit out something that looked impressive, and now they’re selling you a “system.”
Yet ask them what a deal slipping in the last week of the quarter feels like, or why a rep ghosts a perfectly good process, and you get silence. They learned what selling is from an LLM. I’m skeptical of anyone preaching something they’ve never had to live with the consequences of.
Why this matters for you
Because you’ll waste weeks copying their setup and wonder why it doesn’t stick.
I see it on my own team, with people i am talking.
A rep gets excited about a tool, spends two afternoons “experimenting,” and at the end has… a slightly faster way of doing something that was never the bottleneck. They were playing. And playing is fun… it feels like progress, but there’s rarely an outcome at the end. The focus was on the tool, not on how they were already working.
That’s the trap. A good tool can only make a bad process faster. A good process makes almost any tool more useful. Get that order wrong and you’ve automated the mess.
Deloitte’s 2026 Tech Trends report found only 11% of organizations have AI agents actually in production, while 38% are stuck piloting them, the gap between “we’re playing with it” and “it runs our work.” Gartner has long pegged automation failure at roughly 80% when the underlying process isn’t redesigned first. Michael Hammer made the same point in Harvard Business Review back in 1990, and it’s aged frighteningly well: “Don’t automate, obliterate.”
Meaning: fix the process, don’t just bolt a tool onto it.
What I do instead, start with time levers
When I bring AI into how my team works, I don’t start with the shiniest use case. I start with the most repetitive, lowest-judgment task that eats time. Right now that’s reporting.
Reporting is perfect because it’s high-volume, predictable, and already produces data. So I push it to the maximum: the goal is minimum human interaction. Not “AI helps me write the report faster.” More like “the report assembles itself and I review it.”
Here’s the order I follow — process first, tool second:
Pick one repetitive, well-defined step. Reporting, follow-up drafting, CRM cleanup, call summaries. If it varies wildly case-by-case, skip it — standardize the process before you automate it.
Measure the current state. How long does it take today? How many handoffs? Where do errors creep in? Without a baseline you can’t tell if the tool actually helped or just felt cool.
Then point a tool at that exact step. Not the whole workflow. One step. Whether that’s Claude Code, an N8N flow, or a saved prompt matters far less than whether the step was clearly defined first.
Push it toward minimum interaction. Don’t settle for “20% faster.” Ask what it would take for this to run with you only checking the output. That’s where the real time saving lives.
Check the result, then expand — only if it earned it. If the pilot genuinely saved time or caught errors, widen it. If not, kill it. No sunk-cost loyalty to a tool.
The reason this beats tool-chasing: you can swap the tool out next month without rebuilding your way of working. The process is the durable asset. The tool is rented.
This is also why two reps get wildly different results from the same setup — something I dug into a few editions back. The one with a documented process gets leverage. The one treating the LLM like a search box gets frustrated.
⚡ QUICK WIN
Problem: You keep “experimenting” with AI and never reach a clean outcome.
Fix — the One Step Rule, ~10 minutes:
Write down one task you do every week that’s repetitive and rule-based (weekly pipeline summary, recurring follow-up, deal-stage update).
Next to it, write the current time cost and the one thing that makes it annoying.
Give an LLM only that step — with your real example pasted in — and ask: “What would it take to run this with me only reviewing the output, not creating it?”
Why it works: You’re forcing yourself to define the process before reaching for the tool. The constraint (”minimum interaction”) does the thinking for you — it pushes past “slightly faster” toward “barely touches my hands.” One well-defined step beats ten half-played experiments.
📚 READING/LISTENING
(Thumbnail is sketchy but the content is gold!)
“AI Won’t Fix What’s Already Broken: 3 Things to Do Before You Automate Anything” — AI Queens podcast, with Rachel Njiru (~39 min) The whole episode is this edition’s thesis in audio. Njiru’s line that AI amplifies everything — the good and the broken — is the cleanest way I’ve heard it put. Jump to the current-state audit segment (~middle third) and the job vs. task distinction. That second one matters for us: most reps try to automate a whole job (”do my reporting”) when they should be automating one task (”pull these five numbers into this format”). Different industry than ours, same trap.
📄 Michael Hammer, “Reengineering Work: Don’t Automate, Obliterate”
Harvard Business Review, 1990 Yes, 1990. Read it anyway. It’s the original argument that you don’t speed up a broken process, you rethink whether it should exist. Swap “computers” for “AI” and it reads like it was written last week. The contrarian root of everything I said above.
📬 NEW READERS (non-subscribers only)
Not a subscriber? Here’s what you missed recently:
Why your team gets different results from the same AI tools — https://aisaleslab.substack.com/p/why-your-team-gets-different-results
Nobody’s an AI expert. Stop pretending and build systems instead. — https://aisaleslab.substack.com/p/nobodys-an-ai-expert-stop-pretending
Replace your CRM stages with 6 questions that expose real deal health — https://aisaleslab.substack.com/p/replace-your-crm-stages-with-6-questions
Pick one repetitive task this week and define the process before you touch a tool. Then push it toward minimum interaction.
Here’s my question for you — hit reply or drop it in the comments: What’s the last AI tool you “experimented” with that never turned into an actual outcome? What was missing — the tool, or the process underneath it?
Florian
P.S. I’m rebuilding my team’s reporting so it assembles itself and we just review. When it’s running, I’ll show you the exact before-and-after — including the parts that broke.