A friend WhatsApped me Monday morning. 08:28 CEST. He's been running his own AI stack longer than I have. Three weeks into his current ai automation strategy he's convinced he's just clicking buttons faster.
His exact words came in three short bursts:
> "where I'm 'stuck' is that the ai tools are great for research and telling me what to do, but I still have to click the buttons etc" > > "so I'm actually just increasing my workload" > > "are you getting past that?"
I sat on that for a minute. Closed the chat, made coffee, came back and answered him honestly.
> "am i getting stuck on ai actually automating anything? i haven't really been stuck, but i haven't really automated anything but daily logs to check things, go over them and then ship. i have automated an accounting run for a client, that's automated, but it was pretty intense building it and we are testing in sandbox this week. but i'm building my openclaw assistant this week. it seems like that is what you're asking."
Two things on my actual list. Out of four weeks of work, dozens of so-called AI workflows, a vault full of skills and prompts, and a daily build-in-public commitment. Two.
The louder question being asked in 2026 is the same. Is any of this actually automating anything of value, or are we all just clicking buttons faster?
What I would call genuinely automated
Daily logs. A scheduled job runs at 06:06 CEST every morning (give the link to that). It pulls the last 24 hours of Gmail, Google Calendar, Granola transcripts, Cloudflare logs, GitHub commits, and the GiveReady donation digest. It writes a one-page summary into my vault. I open it, read it, edit two or three lines, and ship it. Total time on my side: six to twelve minutes a day. Before the job existed I spent thirty-five to fifty minutes doing the same compilation by hand. That is roughly two hundred hours back across a year if it keeps holding.
A bank reconciliation pipeline for an enterprise client. Six weeks of build. Three thirty-minute calls with the CFO. One Loom walking her actual workflow inside the accounting platform. Fifty-six regression tests passing in 1.5 seconds. Byte-identical reproduction of three months of her hand-built CSV. Sandbox testing this week. A senior engineer reads every line of code before any of it touches a live month-end. The full build write-up went out yesterday in Day 45. When it goes live (call it June at the earliest), it removes roughly twenty hours a month from her plate. Twelve months of that is two hundred and forty hours.
That's it. Two things. Roughly four hundred and forty hours of human time off the table in a year if both hold. Everything else in my day is me clicking buttons. Faster. Still me.
What I would not call automated
Drafting these posts. The AI helps me think, gives me the first five hundred words, catches a cliché I would have missed. I still write the actual post. Three to five hours of my time in a tab with Claude open.
Replying to client emails. Same pattern. AI drafts, I edit, I send.
Outbound prospecting. Apollo is sitting on a Week 4+ pause precisely because the AI was making me faster at the wrong thing.
Reviewing code from a senior engineer. AI-assisted, but the AI is not the reviewer. I'm reading every diff alongside him.
GiveReady's daily improvements. AI drafts the change, I review the diff, I ship. Faster than a year ago. Not automated.
Christopher Penn's 5Ps and why they matter
Christopher Penn writes about marketing AI and his 5P framework is the discipline filter I keep coming back to. Five letters. Purpose, People, Process, Platform, Performance. If you can't write each one down, in one or two sentences, against the thing you're trying to automate, you do not have an automation. You have a thing that tells you what to click.
Purpose. Write down what done looks like, in one sentence, before you build anything. For the bank reconciliation: "Produce a CSV byte-identical to the CFO's hand-built April file, every month, with no manual matching." For the daily log: "A one-page summary of the last 24 hours, in my inbox by 06:30 CEST, with the three numbers I actually look at." If you can't write the sentence, you can't automate the thing.
People. Name the human who has to sign off, every cycle. For the bank rec, the CFO approves every transaction before it posts. For the daily log, me, every morning. For an auto-WordPress posting workflow, someone who would catch a hallucination and not be embarrassed by it. If the answer is "the AI signs off on itself," you do not have an automation, you have a liability.
Process. Render the workflow before writing a line of code. I did not write a single Python file for the bank rec until I had watched a Loom of the CFO doing the manual workflow and could redraw it in markdown as a step-by-step. If the process diagram is fuzzy, the code will be fuzzier.
Platform. Pick one. Stop chasing the next shiny tool. I lost the better part of December and January testing six different AI-assistant tools before settling on the two I'm now running. The cost of that drift was three months of build time I do not get back. Day 38 is the full version of that lesson.
Performance. How do you know it worked? For the bank rec: fifty-six regression tests, byte-identical match to the truth file, zero false postings in sandbox. For the daily log: I open it before 07:00 CEST and the three numbers are correct. For an auto-WordPress workflow: time-on-page, bounce rate, do search engines start punishing the site for thin content. If you can't name the measurement, you are not measuring.
Back to the question, are you automating anything with AI?
He raised the brand-damage worry in the same thread. His words: "if I didn't care, was just shipping anything, I think it would be alright. eg: affiliate site I didn't care about, then totally fine. I'm wrestling with brand damage possibility if we just go full steam ahead (eg auto wordpress posts, I think could work)."
The 5P answer is: auto WordPress posts will damage your brand if you have not named the five Ps, in writing, against the thing you're trying to ship. Auto WordPress posts will not damage your brand if you have, and they have a human signing off, and the performance metric is one you actually check.
Brand-damaging version: "I plugged Claude into WordPress, told it to write three posts a day, walked away."
Value-adding version: I named the purpose (rank for these twenty long-tail queries that match what readers actually search for). I named the people (I read every draft before it ships, and one trusted reader reviews the brand voice once a week). I rendered the process (Claude drafts, I edit for voice, the cliché blacklist runs as a pre-flight, I ship). I picked the platform (Claude plus my voice guide and the cliché blacklist, full stop). I named the performance metric (search impressions on those twenty queries, weekly, plus a manual read of the last seven posts every Sunday).
The first version takes two hours to set up. The second takes two weeks. The first will probably damage the brand. The second will probably move the search needle.
What I'm trying to ship this week
The openclaw assistant. A locally-run AI employee on a small VPS, doing the morning scheduling and Gmail triage. If I can get the 5Ps named cleanly against it (purpose, people, process, platform, performance) and ship a working version this week, that's a third thing on the actually-automated list. If I can't, it gets cut, and I run another week of clicking buttons.
Most AI tools are research assistants with a chat box. They tell you what to do. They are not doing it for you. Calling that "automation" is how you end up at week three convinced you have made things worse, which is roughly where Mitch was at 08:28 Monday morning.
Two things in six weeks, with one more due by Friday. Anything I cannot run through the 5Ps cleanly, in writing, does not ship. Anything that does ship gets a human signature on the way out the door.
Visit finnwardman.com if you want to see what the daily-log automation funds on the WEF side: a 2026 grant round for young people aged 18–26.
Monthly Revenues $11,800 | Clients 2 | Prospects 1, a new one! | Employees - Still just me. For now.
Day 46 of 365.