Alex (@AlexZerbach) header
Alex (@AlexZerbach) profile picture

Previously @ https://teapawt.net/t.co/QmvXrZpGxP - now hacking growth at https://teapawt.net/t.co/3hktJrwXRe

Ridgefield, WA
Joined February 2019
Photo by Alex (@AlexZerbach)Photo by Alex (@AlexZerbach)Photo by Alex (@AlexZerbach)Photo by Alex (@AlexZerbach)Photo by Alex (@AlexZerbach)Photo by Alex (@AlexZerbach)Photo by Alex (@AlexZerbach)Photo by Alex (@AlexZerbach)Photo by Alex (@AlexZerbach)Photo by Alex (@AlexZerbach)
A client's AI security agent blocked our setup on a kickoff call. Our own onboarding got flagged as a threat lol And it was right to. It proposed a two-week, read-only trial, with manual approval for anything that writes or deletes. Better rollout plan than most humans write.
13
Went in to automate a finance team's daily report. Ended up leaving it alone. Turns out it wasn't the problem. They'd already made it pretty efficient. What wrecked their day was the one-off "hey, can you pull this real quick?" asks. So that's what the AI learns first.
28
Saw this on a client team last week. Everyone built their own AI helper for quoting, and each one quoted a little differently. Same company, several versions of the truth. The fix was one shared bot for the core process. Personal bots stay personal. Anyone else hit this?
1
1
115
A client wanted AI to clean up their messy project boards. Our first move wasn't the AI. It was writing down what a clean board actually looks like, in a short SOP. Now the agent compares the real boards to that doc and proposes fixes. Hard to fix a mess nobody's defined.
15
People are tired of hearing about AI. Nobody's tired of getting their Friday afternoon back.
7
Spent a good chunk of a client call stuck on one software connection. We dug through every API setting looking for an ID. It was sitting in the browser's address bar the whole time 🙃 Honestly, half of "AI integration" work is knowing where software hides its stuff.
1
16
Found an AI agent last week checking an inbox every 5 minutes. All day, every day. It was basically the whole reason the owner's AI bill looked scary.
1
18
His answer was fair. His clients don't care, but his team does, because that's how they kick off work and get paid. So I didn't pick for him. I showed him how to audit what each check costs, and he's deciding the speed vs. cost tradeoff himself.
1
3
Turns out with agents, "how fast does this need to happen?" is a pricing question, not a tech question. Nobody asks it until the bill shows up.
Two clients asked me the same thing this week, and it wasn't "what can the AI do?" It was "what happens to our data if you get bought or go under?" Honestly the best question in AI buying right now. If the answer isn't "you own it and can export it anytime," I'd walk.
11
Skills vs routines. I explained this on an onboarding call last week and it finally clicked for people, so here's the plain-English version 🧵
1
1
68
The AI does the grunt work: pulls the info, sizes it up, writes the draft. You still make the call. That's the part of the job you keep, and honestly the fun part.
1
7
Order matters though. Put a schedule on a process nobody wrote down and you just get chaos on time. Skill first, then routine. What's the first SOP you'd hand off?
5
Explained AI agents to a team twice last week and still lost the room 😅 Their feedback was "this feels too theoretical." Fair enough. So I built their weekly prep doc as an automation, live on the call. Questions stopped pretty fast. Turns out nobody buys AI off a slide deck.
3
1
2
98
Hot take after months of putting AI agents into small businesses: most "AI readiness" is just API readiness. The agent is rarely what breaks. The old software running the business is. 🧵
5
1
67
Before you spend a dollar on agents, answer 3 questions: 1. Can my core system expose or export data? (API, webhook, even a CSV) 2. Who owns the admin login and permissions? 3. Does any login need 2FA from someone's phone? No, no, yes? Fix the stack first.
1
4
Not sexy advice. But it's the cheapest AI consulting you'll get this year.
2