Q6: survival plays.
Rate limit. Degrade. Queue. Cache.
System can slow. Can't fully collapse.
Performance design isn't engineering jargon — it's asking these real-world questions before you ship.
Template ↓ for your next AI build.
Learn AI through comics — I'm Badbot.
Q4: how long should it take?
For AI especially: time-to-first-token > total time.
If something's streaming, no one feels stuck.
Q5: how many will fail under load?
Things break — network, API limits, model timeouts.
Build a UX fallback. Not an infinite spinner.
Your vibe-coded site looked perfect in demo.
Share it with friends → it crashes in 3 seconds.
AI didn't fail you.
You just never told it the real-world scenario.
That blind spot has a name: Performance Design 🧵
The hard part of ToB AI isn't tech.
It's embedding AI into human companies.
If you're a PM, the most valuable skill for the next 3 years is translation — turning clients' dirty biz problems into things AI can actually do.
Learn AI through comics — I'm Badbot. (4/4)
Why is everyone going hands-on?
Two reasons:
1. The Demo→business gap is huge AND uniquely shaped per company. Generic templates won't cut it.
2. Finance math: $1 of software = $6 of services. MIT: 95% of enterprise GenAI pilots stay stuck. (3/4)
$4B Deployment Co. $1.5B services arm. 150 field engineers shipped to client sites.
That's OpenAI and Anthropic now — not API kings, but consulting houses with stickers.
Why is selling AI getting this dirty? A thread (1/4):
This is huge for AI Agents. Lower memory cost means models can handle much bigger workspaces.
DeepSeek V4's secret? It didn't try to memorize more — it changed how it reads.
Follow @badbot_X for more AI comics.
Learn AI through comics — I'm Badbot
The genius part: V4 alternates CSA and HCA across model layers.
Distant info → compressed into index
Big picture → compressed into outline
Nearby info → kept as-is
Just like how humans switch between close reading and skimming. Elegant.
AI now writes PRDs and reports better than ever.
But your boss can spot 'AI flavor' in 1 second.
5 tricks I use to humanize AI output — so it actually looks like ME wrote it 🧵
The real shift: humanizing AI = using YOUR experience and taste to cut what shouldn't belong to you.
I built a 6-step Skill that runs before every report or PRD.
Stay sharp, stay human.
Follow @badbot_X for more AI comics 🤖
Tip 4: Rank info, break structure
AI loves the rule of three (hi, ChatGPT). Cut filler. Reset the doc's center of gravity.
Tip 5: Re-narrate
Charts + your wording > AI's wording. Make YOUR judgment visible.
From poetry to mythology to pure pragmatism — each company's naming tells a story.
What does your favorite AI name say about its maker?
Follow @BadbotAI_ for visual AI explainers
Other companies' naming reveals their AI philosophy:
Claude — after Claude Shannon, info theory's father. Humanities & tribute.
Gemini — stars & twins. Two teams merged. Next frontier.
ChatGPT — function = name. No metaphor, pure efficiency.
Anthropic accidentally leaked their most powerful model ever — Claude Mythos.
But here's the fun part: every Claude model is named after a literary form.
Let's decode what each name means 🧵
