comicexplain
explained comic · 21 pages · your annotated copy
an educational manga explainer
cover artwork
Free by Design: Apple's iOS 27 AI Stack and the Zero-Cost Inference Frontier
comicexplain
session notes
kept private to your copy — sign in to add notes.
about this comic
Free by Design: Apple's iOS 27 AI Stack and the Zero-Cost Inference Frontier
comic page 1
my take on page 1
key idea
Most AI work can move off the paid cloud onto free Apple paths.
flip ↺
why this matters
Understanding the four paths lets developers cut inference bills without losing capability.
flip ↺
p. 1 · 21
comic page 2
my take on page 2
key idea
Most cloud spend comes from calls that shouldn't be cloud calls.
flip ↺
why this matters
Recognizing waste is the first step to cutting an AI app's largest recurring expense.
flip ↺
p. 2 · 21
comic page 3
my take on page 3
key idea
Know the four paths and their costs before designing anything.
flip ↺
why this matters
The whole cost strategy depends on matching each task to the right path.
flip ↺
p. 3 · 21
comic page 4
my take on page 4
key idea
The on-device model is a free, always-available local worker.
flip ↺
why this matters
It replaces many paid cloud calls with zero-cost local ones.
flip ↺
p. 4 · 21
comic page 5
my take on page 5
key idea
Tiny context makes it a worker, not the whole intelligence.
flip ↺
why this matters
Knowing the limit prevents designing tasks the local model can't hold.
flip ↺
p. 5 · 21
comic page 6
my take on page 6
key idea
Bring your own models and run them free on-device.
flip ↺
why this matters
It gives developers control and cost-free specialized inference.
flip ↺
p. 6 · 21
comic page 7
my take on page 7
key idea
Core AI is an efficient purpose-built on-device inference runtime.
flip ↺
why this matters
These features let large specialized models run practically on a phone.
flip ↺
p. 7 · 21
comic page 8
my take on page 8
key idea
One common interface lets you swap underlying models freely.
flip ↺
why this matters
Shared orchestration avoids duplicate code for local and cloud models.
flip ↺
p. 8 · 21
comic page 9
my take on page 9
key idea
A session can adapt its model and tools on the fly.
flip ↺
why this matters
It supports escalation and specialization without breaking the conversation.
flip ↺
p. 9 · 21
comic page 10
my take on page 10
key idea
PCC adds reasoning and context, free for eligible small apps.
flip ↺
why this matters
It gives indie developers subsidized access to stronger cloud inference.
flip ↺
p. 10 · 21
comic page 11
my take on page 11
key idea
Eligible indie apps get subsidized cloud reasoning at no cost.
flip ↺
why this matters
It lowers the barrier for small teams to ship capable AI features.
flip ↺
p. 11 · 21
comic page 12
my take on page 12
key idea
Offload cheap calls locally to shrink the paid cloud bill.
flip ↺
why this matters
It reframes cost control as routing, not just picking a cheaper model.
flip ↺
p. 12 · 21
comic page 13
my take on page 13
key idea
On-device translation eliminates a paid cloud service.
flip ↺
why this matters
Multilingual apps can drop another recurring cloud cost.
flip ↺
p. 13 · 21
comic page 14
my take on page 14
key idea
Local OCR plus local model keeps documents on the phone for free.
flip ↺
why this matters
It cuts cost and improves privacy for document workflows.
flip ↺
p. 14 · 21
comic page 15
my take on page 15
key idea
System-provided intelligence handles text and image tasks free.
flip ↺
why this matters
More common features can rely on built-in models instead of paid APIs.
flip ↺
p. 15 · 21
comic page 16
my take on page 16
key idea
Programmatic Apple image generation is gone in iOS 27.
flip ↺
why this matters
Apps built on iOS 26 image APIs must migrate before public release.
flip ↺
p. 16 · 21
comic page 17
my take on page 17
key idea
Match each task to its cheapest capable path.
flip ↺
why this matters
A clear routing table turns cost savings into a repeatable design rule.
flip ↺
p. 17 · 21
comic page 18
my take on page 18
key idea
Escalate to paid models only as a final fallback.
flip ↺
why this matters
Layered escalation delivers capability while minimizing paid calls.
flip ↺
p. 18 · 21
comic page 19
my take on page 19
key idea
Use the local model for narrow tasks, not world knowledge.
flip ↺
why this matters
Matching tasks to its strengths removes many paid calls safely.
flip ↺
p. 19 · 21
comic page 20
my take on page 20
key idea
Ask which pipeline parts can become free, not how smart the model is.
flip ↺
why this matters
That question directs effort toward the largest achievable cost savings.
flip ↺
p. 20 · 21
comic page 21
my take on page 21
key idea
Treat inference cost as routing, then measure, profile, and question the limits.
flip ↺
why this matters
It turns the comic's lesson into concrete engineering investigation and honest caveats.
flip ↺
p. 21 · 21
flashcards · all 21
key idea
Most AI work can move off the paid cloud onto free Apple paths.
click to flip ↺
why this matters
Understanding the four paths lets developers cut inference bills without losing capability.
click to flip ↺
card 1 · 21
one card per page — cycle through before you close the book.
index of key ideas
1Most AI work can move off the paid cloud onto free Apple paths.
2Most cloud spend comes from calls that shouldn't be cloud calls.
3Know the four paths and their costs before designing anything.
4The on-device model is a free, always-available local worker.
5Tiny context makes it a worker, not the whole intelligence.
6Bring your own models and run them free on-device.
7Core AI is an efficient purpose-built on-device inference runtime.
8One common interface lets you swap underlying models freely.
9A session can adapt its model and tools on the fly.
10PCC adds reasoning and context, free for eligible small apps.
11Eligible indie apps get subsidized cloud reasoning at no cost.
12Offload cheap calls locally to shrink the paid cloud bill.
13On-device translation eliminates a paid cloud service.
14Local OCR plus local model keeps documents on the phone for free.
15System-provided intelligence handles text and image tasks free.
16Programmatic Apple image generation is gone in iOS 27.
17Match each task to its cheapest capable path.
18Escalate to paid models only as a final fallback.
19Use the local model for narrow tasks, not world knowledge.
20Ask which pipeline parts can become free, not how smart the model is.
21Treat inference cost as routing, then measure, profile, and question the limits.
comicexplain
paste a paper, get an annotated comic — pages appear in seconds.
front cover