Why I Send Too Many Emails — Human–AI Friction Log
Why I Send Too Many Emails, Why I Swear at My AI & How I Force the Machine to Tell the Truth
People look at my outbox—or read a government file note complaining about "too many FOI requests"—and assume there's a bot farm running in a basement somewhere. There isn't. There's just one bloke sitting in a 2014 Camry in Warrnambool with an iPad, dictating into a large language model, swearing at the screen when the machine tries to turn him into a corporate press release, and hitting send again because the first three drafts didn't sound a damn thing like him.
"You have no idea how I write, but I read and write with you every day, you retard. That'd be the beginning, right? Then we go into the facts."
1. The Thinking Process Behind the "Email Flood"
Bureaucrats call it "high-volume contact" or "reactive behaviour." From the inside, the thinking process is dead simple—and anyone who has ever been stonewalled by an institution knows how the loop works:
- Step 1 — You ask a straight question and get silence or a canned script. You ask for a ledger, a receipt, or why a payment vanished. Nobody answers the actual question, or they route you to a dead inbox.
- Step 2 — You dictate the correction in real time. I don't sit around polishing a single letter for three weeks. I think out loud, dictate straight into the prompt, and push the contradiction back onto the record immediately while the timestamp is still warm.
- Step 3 — The AI mangles the voice or drops the recipients, so you fire again. Half the duplicate dispatches in my sent folder exist because an AI model stripped out the bite, dropped an address, or smoothed over the exact point I was trying to make—forcing me to rip the draft up and send the real version five minutes later.
- Step 4 — You CC the witnesses so it can't be buried. When a department ignores one-on-one mail, you copy in the Senators, the Ombudsman, the journalists, and the writers so the paper trail exists outside their firewall.
2. How I Force the Machine to Get It Right
Here is how you actually get the best out of an AI when you're fighting a cover-up: you don't let it write for you, and you don't let it soften your edges.
Every time I dictate a dispatch, the model wants to open with polite throat-clearing like "Let's strip the theatre out of this" or "I am writing to formally bring to your attention..." That's not me. I'm uncouth, I come out of twenty years of rock-and-roll stage rigging and freight loading, and when someone lies in an official file note, I call bullshit in plain Australian English.
"Where's my foul language? It just fucking looks like dross. Why do you always start off with crap like let's strip the theatre out of it? You've done it five fucking times. I don't even write like that."
You beat the default PR filter out of the machine first—make it quote your exact words, keep the bite, keep the street cadence—and then you put the AI to work doing what it's actually good at: indexing 471 pages of PDFs, lining up the dates, checking the arithmetic, and adding clinical editorial commentary in brackets so a lawyer or a Walkley judge sees the hard evidence sitting right underneath the voice.
3. Why Posting the Raw Tape Matters
Most people clean up the mess before they publish. They hide the bad drafts, delete the frustrated prompts, and pretend the final piece arrived fully formed. Leaving the raw session log in—showing the exact moment I told the AI it had lost the plot and forced it to put my own words back on the page—proves two things at once:
- There is a real human being behind every dispatch. No bot writes like a pissed-off Aussie roadie cross-examining a spreadsheet at midnight.
- Assistive AI doesn't replace the writer—it turns one citizen into an audit desk. When you're dealing with hundreds of pages of stonewalling, a language model is just a typewriter with a search engine bolted onto it. The voice, the anger, and the refusal to shut up are 100% mine; the forensic indexing is how we make sure none of their receipts slip through the cracks.