Apparently, my niche is being interested in too many things. 😄
tech, football, running, cycling, coffee, painting, music…
we’ll see what shows up here tomorrow.
72
I spent a week testing post times.
The clock was not the bottleneck. I was writing conclusions, not conversations.
So I am changing one thing: one concrete incident, then a genuine question.
When did a buyer last change how you explained your product?
48
This week one of my automations refused a scheduled task. It could not verify the screen state it needed, so it stopped and flagged me instead of guessing.
Cost: one day of delay, zero cleanup.
A system that stops when blind beats one that acts on assumptions.
33
Founders don't buy AI.
They buy fewer dropped balls:
missed follow-ups
failed handoffs
silent errors
unclear ownership
If an agent can't improve a real workflow, it's a toy with an invoice template.
19
The first AI automation question I ask is not, "Which model should we use?"
It is: which workflow is repetitive, measurable, expensive when it fails, and safe to contain inside a small pilot?
If that answer is vague, adding AI usually makes the confusion faster.
12
AI agents rarely fail because the model is too weak.
They fail because nobody designed:
• ownership
• handoffs
• retries
• audit trails
• the moment a human takes over
The demo is the easy part. Operations is the product.
15
Building AI agents is easy until billing, support, and edge cases show up.
Founder @onivordigital · 20y ops · shipping from Indonesia.
I write about the gap between demo and business.
→ onivor.com
2
132
Fable 5 is a reminder that model capability is becoming a product design problem, not only a benchmark problem.
The hard part for builders is deciding when to trust the strongest model, when to route down, and how much of that decision should be visible to the user.
120
I've been thinking about the ‘AI can build a business in 15 minutes’ narrative.
The prototype may take 15 minutes.
The business does not.
Distribution, edge cases, support, billing, trust, and judgment still take real operating discipline.
69
Saw another "I built this in a day with AI" post.
No shade. But: shipping fast in a day is not the flex anymore. The flex is shipping, waiting 48hrs, and having the stomach to delete what didn't land.
1
46
Opened my agent dashboard this morning. Three of them made reasonable mistakes overnight. One was funny (hallucinated a product feature with a creative name). Two were subtle (slightly wrong summaries that would have compounded). I caught it because I looked. The judgment is still the job.
23
My AI founder loop looks like this now: Idea → 4 agents draft options → human picks the least embarrassing one → user signal decides if we ship another round. Still miss. Still iterate. But the cycle that used to take a week now takes a day. The taste is still mine. The speed is borrowed.
18
Founder judgment under AI leverage: the boundary isn't 'should I use AI?' It's knowing when the agent output is good enough vs when it needs a human loop. I'm wrong about this 20% of the time. The wins are from catching the other 80% faster, not being right 100%.
11
$2M in tokens sounds like cheat codes until you're debugging why your agent burned through half of it on a recursive loop at 2am. The real unlock isn't the tokens. It's knowing which workflows deserve the spend vs which ones just feel impressive to automate.
i am excited to see what will happen with tokenmaxxing startups, both for how they work internally and the products they can build.
openai offered to invest $2M in tokens into every startup in the current yc batch.
happy building!
15
New thing I learned running 12 agents: the failures are never dramatic. They're small judgment errors that compound. Agent A summarizes slightly wrong. Agent B builds on that summary. Agent C ships based on B's output. By the time I catch it, three decisions deep, the cleanup is worse than doing it manually. Now I pause at every handoff. Annoying? Yes. But the "fast" path was actually just a prettier way to create debt.
10
My AI founder loop:
Idea → agent drafts → script checks → human judges → agent fixes → user signal decides.
Still messy. Still breaks. Sometimes funny.
But vs doing it manually, it feels like building a small company operating system one workflow at a time.
1
1
23
AI makes solo founders look bigger than they are.
That is useful, but also dangerous.
You can generate roadmaps, code, copy, research, and support replies very fast now.
But if the judgment is weak, you are not building faster. You are just creating better-looking confusion.
4
AI makes the build cycle shorter, but it does not remove taste.
You can launch faster now. You can also launch the wrong thing faster.
The advantage is not “I used AI.”
The advantage is learning faster than your confusion compounds 😄
6

