Yesterday morning, I had never used Claude Code. By the end of the day, I had a web crawler running on a scheduled cron job, a beefed-up, 99-page site audit that surfaced findings the previous version missed entirely, and an automated pipeline that will produce a Word document audit report every Monday morning without me again touching a terminal.
I am not a developer.
I run a two-person association marketing department. And the honest lesson from this experience is not that I did something technically impressive. It is that the barrier to this kind of work is lower than most small-staff professionals assume — and the return on clearing it is significant enough to be worth the discomfort of trying.
What the Problem Actually Was
For months, the site audit workflow looked like this: I would generate a crawler script in Claude chat, copy it to my machine, run it in Terminal, paste the results back into chat, and ask for the audit report. That process worked. It produced real findings and real deliverables. But it had a ceiling.
Every iteration required me as the relay. I would run the script, share the error, wait for a fix, run it again. I would upload the output, describe what I was seeing, wait for the analysis. I was the connective tissue between the tool and the result — and that meant every improvement in the output cost me time proportional to the improvement.
Claude Code removed the relay.
It runs the script itself, reads the output directly, identifies the errors, fixes them, and runs again — without waiting for me to copy and paste anything. What took a multi-session cycle of generate, run, report, fix now happens in a single continuous work session.
The barrier to this kind of work is lower than most small-staff professionals assume — and the return on clearing it is significant enough to be worth the discomfort of trying.
What One Day Actually Produced
Starting from installation, here is what happened in roughly one working day:
The crawler was rebuilt from scratch. The previous version captured six data fields per page. Claude Code rewrote it to capture 13 — adding meta descriptions, H1 and H2 heading structure, word count, calls to action, images missing alt text, and redirect detection. That is not a minor upgrade. It is a materially different audit instrument.
A cron job was configured to run the crawler automatically every Monday at 7:45am. The Mac wakes itself at that time, runs the crawl, and saves the output to a dated log file. No manual trigger required.
The 99-page full-site crawl surfaced findings that previous runs missed:
- A publicly visible 403 error on a page still in the site navigation.
- The home page — the highest-traffic page on the site — with zero H1 tags.
- A fundraising page with seven images missing alt text.
- A membership count on a public advocacy page that was one update behind.
The audit report built itself into a Word document. Same format as the previous audit cycles. Severity badges, executive summary scorecard, 99-page table, prioritized action list. Delivered without me formatting anything.
Why This Matters for Small Staff
The association sector runs on lean teams. Two-person marketing departments are not unusual. Three-person total staff operations are common. In that environment, the limiting factor is almost never skill — it is capacity. There is simply not enough time to do the work that needs doing and also build the systems that would make doing it easier.
AI does not solve that problem by adding headcount. It solves it by removing you from the parts of the process that do not require your judgment.
Running a script does not require your judgment. Copying output from a terminal window does not require your judgment. Reformatting a report does not require your judgment. What requires your judgment is knowing what to look for, deciding what matters, and determining what to do about it.
Everything else is relay work, and relay work is exactly what AI handles best.
When the relay work is automated, the professional gets back the capacity that was being consumed by it. That capacity goes toward the thinking work. And the thinking work is where the organizational value actually lives.
The Honest Account of How Hard It Was
The setup was not seamless.
Claude Code had to be installed, and the PATH configuration required two attempts because macOS defaults to zsh and the first fix targeted bash. The cron job needed a caffeinate flag to keep the Mac awake through a 30-to-45-minute crawl. The npm dependencies had to be installed before the job would run. Each of those things produced an error before it produced a fix.
None of those errors required me to understand what was happening at a technical level. They required me to describe what I was seeing and trust the process. That is a different skill than terminal fluency, and it is a skill that association professionals already have.
We describe problems clearly. We communicate what we need. We recognize when an output matches what we asked for and when it does not. Those are the skills that make AI adoption work, and they are not technical skills.
The discomfort is real.
Opening a terminal window for the first time and running commands you do not fully understand is uncomfortable in a specific way. It feels like operating without a net. What I can tell you from yesterday is that the net exists — it is the AI walking you through every step — and the discomfort clears faster than you expect once the first thing works.
What the Repeatable Principle Is
The specific workflow — a web crawler, a cron job, a site audit — is not the point. The principle behind it is.
Find the work in your operation that is iterative, file-heavy, and currently requires you as the relay between two tools. That is where AI augmentation produces the clearest return. The newsletter that gets reformatted every month. The membership report that gets rebuilt from the same data pull every quarter. The conference communications checklist that gets recreated from memory every cycle. The process documentation that should exist but has never been written because writing it was slower than just doing the thing again.
Each of those is a relay problem. Each of them has a version where AI handles the relay and you handle the judgment. Getting there requires a willingness to spend a working day — or in many cases, a working hour — building the system instead of doing the task.
Small-staff associations cannot afford to hire their way out of capacity problems. Most of them know this. What is less widely understood is that the alternative to hiring is not just working harder — it is working differently. Building systems that run without you. Automating the relay. Reserving your time for the work that only you can do.
That is what AI makes possible for a two-person department. Not the elimination of the professional. The elevation of what the professional spends their time on.
Yesterday, I spent a working day building a system. Every Monday from here, that system runs itself. That is the math that small-staff association professionals need to start doing.
Kitchen's Ink 
Leave a comment