What I Automated, And What I Didn't

What I Actually Automated (and What I Didn't)

This morning, an intake agent ran at 6am while I was asleep, and by the time I opened my laptop a new client brief was already organized: the ask pulled apart, deliverables listed, a first-pass timeline proposed, all waiting for me to review it over coffee. Absolutely nothing about that sentence used to be true a year ago, and the interesting part isn't the automation, it's the fences around it.

A new client intake form triggers a workflow in n8n. It receives the raw submission from my site and inserts it into a database, sniffs out the project type, budget range, and timeline, and drops it all into a structured brief using the same template I've used on my own for years. It checks the budget against my minimums and flags anything under it instead of silently accepting it. Next, it drafts a first-pass timeline based on the project type and my current calendar. But it stops there. It doesn't send anything directly to the client, and it doesn't commit to a date. It just writes the whole thing into an Apple Notes folder labeled ‘Project Inbox’, ready for me to read before the coffee finishes brewing.

That last part is actual design work. It's easy to overlook because "it stops and waits for you" feels like exactly the opposite of automation- but surprise - it isn't. Deciding where the workflow's authority ends is a human design decision with the same weight as deciding what it should be doing in the first place. This particular intake workflow is allowed to organize, structure, and flag information from the client’s form submission and my planning & pricing structure. It’s not allowed to promise a client anything, and especially neither a timeline or a price, because those are decisions that need my read when considering the entire client relationship, not just the raw numbers on their intake form. That “stop and ask” line didn't get put there as a general rule around AI and trust. It came from looking at the task and asking myself what would happen if it’s done wrong. Is this the communications standard you want to set with your clients? In my opinion client touchpoints are very human, personal connections, and I’d hate to lose a relationship over a misunderstanding that happened while I wasn’t in the loop.

The first version of my agent didn't have a budget-flag step. It just built the timeline and left it at that. And because of that, I nearly sent a project plan to a client whose budget wouldn’t remotely support the scope we’d planned for, because I’d trusted the workflow's output without checking. Fortunately, I caught the budget mismatch before reaching out to the client, but think often of how our working relationship would have been strained had I not. And that's the real failure mode with automation like this: not that it does something obviously wrong, but that it does work that’s just believable enough that you stop watching. The fix wasn't adding more automation, it was adding a simple, explicit check that the workflow has to fail loudly on, instead of silently producing a clean-looking lie.

There's a bigger version of this problem right around the corner once you're juggling more than one agentic workflow at once, cost, drift, who's watching all of these workers and workflows at the same time. That's its own, separate piece I’ll talk about properly soon. But for now, the smaller takeaway holds true: automate your organizing, not your decisions, and place in human-in-the-loop checkpoints at key steps to assure your workflow can't quietly fail, while pretending everything's fine.

What's the one regular part of your day-to-day process you'd hand over to an agent like this first? Drop it in comments or the #creativeai channel on my Discord, I'll help you think through how to automate the jobs that sap your creative energy the most.

And if this raises more questions, please (securely, privately) drop your email into the Creative AI page for a ping when the book that talks about all this goes live. Tom and I would love to get your thoughts on how to ethically manage automating your processes with AI, and build workflows that you actually trust.