You can just install Claude Code tonight

You can install Claude Code tonight. You don’t need a steering committee. Skip the SOC2 review and the enterprise rollout plan. You need an evening, twenty bucks, and one workflow you’d like to stop doing by hand.

That’s the whole post. The rest is just receipts.

Denning said the quiet part out loud

Tim Denning wrote a piece in November about a phrase that’s been floating around for awhile: “you can just do things.” His framing was sharp. “You can just do things works because it prioritizes momentum.” He told the story of a guy named Dave who took an unpaid gig in Malaysia and turned it into six-figure clients despite having no formal credentials. Dave didn’t out-credential anyone. He out-moved them.

The line we keep coming back to: “Normal behavior is forgotten. Only weird behavior survives.”

Six months of “evaluating AI” is normal behavior. Installing Claude Code on a Tuesday night and pointing it at your messiest spreadsheet is the weird behavior. Guess which one shows up in next year’s margin.

What “evaluating” actually looks like

We’ve sat across the table from enough operators to recognize the pattern. The conversation goes something like this:

• “We’ve been looking at AI for about six months.”
• “We’re waiting on guidance from our IT partner.”
• “We want to see how others in our industry are using it first.”
• “We’ve been talking to a few vendors.”

Meanwhile the calendar is stuffed with $15K assessments and pilot scoping calls. Nothing has shipped. Nobody on the team has actually used the tools. The “AI strategy” is a Notion doc.

Then they find out a competitor’s owner downloaded Claude Pro for $20, spent a weekend with it, and now closes month-end books in a day and a half instead of five.

The tools are already cheap

Here are the actual numbers:

• Claude Pro: $20/month. Web interface, no API key required.
• ChatGPT Plus: $20/month. Same.
• Claude Code: Free CLI; you pay for tokens. Most operators we’ve watched start out under $50/month while they figure out what’s useful.
• n8n: Open source. Self-host for the cost of a $6/month droplet, or use the cloud version starting at $20/month.

The total cost to “just try one thing” is roughly the price of a team lunch. Compare that to six months in evaluation mode, the meetings on your calendar, and the $15K-$50K assessment proposals sitting in your inbox.

Two operators, same Tuesday night

Take a hypothetical 40-person commercial plumbing contractor in Richmond. Their controller spends roughly 4 hours a week matching vendor invoices to purchase orders and field tickets — three apps, two spreadsheets, a lot of swearing. Owner installs Claude Code on a Tuesday night, hands it a folder of last month’s PDFs, and asks it to flag mismatches. By Thursday the reconciliation runs in under 30 minutes. It’s a hypothetical, but the workflow is real; we’ve watched it happen with a near-identical setup. About 180 hours a year back to the controller. At a loaded rate of $55/hour, roughly $9,900 in recovered capacity. The owner spent $20.

Now take a 22-person dental practice doing around $4M in revenue — we’ve seen this pattern at a similar-sized practice. The office manager writes the same five patient communication templates ten different ways every week — insurance follow-ups, treatment plan explanations, no-show messages. She drafts them in ChatGPT Plus, refines the prompts over a weekend, and the team standardizes on the outputs. Time on written communication drops from roughly 6 hours a week to under 1. That’s 250 hours a year at the front desk that go toward patients instead of typing. Still $20/month.

No consultant. No 90-day pilot. No org-wide buy-in. Just one operator deciding Tuesday night was the night.

“But we need to be careful”

We hear this constantly, and there’s a legitimate version of it. HIPAA matters. Client data matters. You can’t paste a payroll file into a free chatbot. Those are real constraints.

But careful and paralyzed are two different things. Careful is using Claude’s Business plan, which doesn’t train on your data. Or testing on dummy invoices before real ones. Or running one workflow on your own laptop for a week before involving anyone else.

Denning again: “You don’t learn unless you break things.” You learn what AI can and can’t do for your business by using it on your business. Not by reading another McKinsey report about it.

The head start compounds

We can already tell who tinkered in 2024 and 2025 versus who waited. The tinkerers aren’t smarter or more technical. Half of them couldn’t tell you what an LLM stands for. They just opened the laptop one night, paid $20, and tried one thing. Then another. Then a third.

The operators who started a year ago are running workflows their competitors don’t even know exist. They’ve fired vendors they didn’t need and moved their people onto harder work instead of replacing them. We see their margins running 4-7 points better than peers who waited, and that gap shows up in the P&L every month — quietly, without a press release.

Denning closes his piece with this: “Certainty is a myth. It never existed.” You will never have a complete picture of what AI can do for your business before you start. Nobody does. The operators winning right now made peace with that and started anyway.

The tools are sitting there. Twenty dollars. One evening. One workflow.