Zak Kann - AI Automation for SMBs

03/24/2026

Opus 4.6 is a major gaslighter.

It loves to tell you issues are pre-existing, not caused by its work. "Oh that bug? That was already there before I touched it."

Every testing flow now needs a "verify pre-existing issues by running on the old code" step.

Trust but verify. Especially with AI.

03/23/2026

In 2022, AI struggled to multiply two-digit numbers.
In 2023, it outscored most law school graduates.
In 2024, it was shipping production code and breaking down PhD-level research.
By the end of 2025, elite developers were openly admitting AI handles the bulk of their engineering.
We're barely into 2026, and it already feels like the ground is shifting faster than ever.

03/20/2026

My workday in 2026: describe what I want built, watch 6 AI agents spin up in parallel worktrees, each doing deep research before writing a single line of code.

I went from managing engineers to spawning an unstoppable army of transient wizards.

03/05/2026

I track one number every Monday morning. It's not revenue.

For the longest time I was checking my Stripe dashboard first thing every morning. Revenue up? Good day. Revenue down? Bad day.

Turns out that's a terrible way to run a business.

Revenue is a lagging indicator. By the time it drops, the damage happened 3 weeks ago. By the time it spikes, you made the right call a month back. You're always reacting to old news.

The number I actually track now: activated users in the last 7 days. Not signups — people who signed up AND did the thing my product exists for.

Here's why this one metric runs my entire week:

If activations are up, growth is coming — even if revenue hasn't moved yet. If activations are flat, something is broken in onboarding and I need to fix it before it shows up as churn in 3 weeks. If activations drop, I don't panic about revenue. I go find where people are falling off.

Revenue is the scoreboard. Activations are the game. You can't change the scoreboard, but you can change how you play.

What's the one number you check first every week?

03/04/2026

5 years later, I think this is the most prescient article I ever wrote.

There are no safe jobs, but lifelong learning can make you safer

03/04/2026

The micro SaaS market is projected to hit $60B by 2030.

Everyone's reading that number and thinking scale. I read it and think: calm.

The founders pulling ahead aren't the ones raising Series A. They're the ones running $30k/month businesses with zero employees. 70%+ profit margins. No board meetings. No runway conversations.

I've watched this shift happen in real time. The indie hackers winning in 2026 all have the same playbook:

They picked a niche so specific it sounds boring. They ship alone (or with AI). And they optimized for margin, not growth rate.

Meanwhile the VC-backed competitor is burning $200k/month trying to be everything for everyone.

I'm not anti-ambition. I'm anti-complexity-for-its-own-sake.

The best business I ever built was the one where I woke up on Monday and didn't dread opening Slack. That's not a lifestyle business. That's a competitive advantage.

Chase calm. The money follows.

03/04/2026

Stop building AI wrappers. Start building AI workflows.

I keep seeing the same startup pitch: "We built a UI on top of GPT-4."
Cool. So did 10,000 other people last weekend.

Here's the thing though — the winners in AI aren't building chatbot interfaces. They're embedding AI into workflows that already exist.

Nobody wakes up wanting "an AI tool." They wake up wanting their job to suck less.

The real opportunity?
→ Take a painful, repetitive process someone already does
→ Plug AI into the middle of it
→ Save them 10 hours a week

A wrapper gets replaced the moment OpenAI ships a better playground. A workflow becomes the thing people can't live without.

Build the thing that's hard to rip out, not hard to replicate.

02/18/2026

How long before LLMs are programmed with near real-time data?

02/18/2026

Remember when all us white collar workers felt bad for blue collar workers who were gonna be replaced by robots any day?
That was less than 4 years ago.

02/13/2026

My big AI prediction for 2026 was more prominence of AI orchestrators, bots whose primary purpose was to orchestrate more specialized bots.
It's the only way to solve context clutter and the fine-tuning problem with modern LLMs.

Well, we must be getting closer to the singularity, because what in the past would have been a whole of 2026 prediction apparently took less than a month to come to fruition.

02/12/2026

I still write some code, but I haven't written a Pull Request summary in over a year.
Let AI do what it's good at.

02/11/2026

Me creating a logo:
"Claude, help me come up with a logo idea for my business [description]."
"Gemini, here's my idea. Make it modern and cool."
"ChatGPT, here's my logo prototype. Fix it and make it look professional."

Someday, I'll have to let one of them go.