Share
Embed
Slide 1
Slide 1 text
NEBRASKA.CODE() 2026
AI Loves Design
Patterns
(Too Much)
at
bo
ten
ci
& re
r
y
e
v
o
c
n
o
t
i
d
d
a
y
r
o
t
t
c
a
e
p
d
t
d
u
a
e
a
h
k
t
l
o
a
25-min Spotlight · Intermediate · Software Craftsmanship × AI
Yes, it
d
an r Fa r
Slide 2
Slide 2 text
A I Di s c l a i m er
1
These slides was designed with Claude Design
2
This talk is about AI generated code...that was generated with Co-pilot and
Claude.
3
Idea, wording and research is my own, with full references provided.
w
o
h
e
h
t
t
w
On i h
s
→
Slide 3
Slide 3 text
No content
Slide 4
Slide 4 text
No content
Slide 5
Slide 5 text
E X H I BI T A · G E N ER AT E D BY A I
"Create a user."
←I as
f u r
at . I go
n es.
Se m ke A
or tom
n
l
r
i
a
f
e
s
i
t
e
d
t
e
d
p
p
t
e
r
a
y
t
r
s
a
w
n
p
t
e
i
t
i
n
s
l
o
S
o
i
i
e
s
b
l
p
o
p
y
r
e
r
r
c
e
u
e
o
e
o
r
l
e
r
n
s
C
d
a
e
n
I
t
r
t
o
o
r
a
a
h
f
d
t
y
y
i
e
l
r
r
k
a
s
l
n
i
i
p
s
s
e
o
o
p
p
u
Uh, is 't e EF r 's D e l a y
re to p r , so C ot
p a
re to r u y
ep or ??
-im m
ad h
s at
s
Slide 6
Slide 6 text
E X H I BI T A · G E N ER AT E D BY A I
"Create a user."
Slide 7
Slide 7 text
AI pattern overuse is learnable,
spottable, and fixable.
Slide 8
Slide 8 text
"The genie seems to assume that its
planetary-sized brain is capable of
handling any amount of complexity,
so it needn't ever reduce complexity."
c
e
B
t
— Ken
k
Slide 9
Slide 9 text
Why AI Has a Pattern
Addiction
01
Un e s
d
o
t
d
o
p
s
s
u
s
l
e
h
u
a
c
h
g
n
i
n
a
t
r
d
SECTION ONE OF FIVE
t e
se
p
re
n
it.
Slide 10
Slide 10 text
T H E DI A G N OS I S
Pattern addiction, defined
1
AI is trained on code, "good code"
2
Patterns signal intent in ambiguous prompts
3
The addiction compounds: your codebase is part of the prompt.
a
i
l
m
a
d
n
u
so
f
i r? →
Slide 11
Slide 11 text
“
The probability that AI gets something wrong
can increase with a bad codebase, because it
will pick up on existing patterns.
r
o
d
n
o
L
C
e
l
k
a
t
i
g
— Bir t Böc e r, Q on
n, Ap 2026
Slide 12
Slide 12 text
A 2026 study of Cursor-generated
projects found 91% average functional
correctness
BUT....
alongside 1,305 CodeScene and 3,193
SonarQube design issues across 10
projects
Correctness benchmarks can hide design quality
GitClear's multi-year study of 211M
changed lines found copy-paste code
rose from 8.3% to 12.3% while
refactoring commits fell from 25% to
under 10% as AI assistance grew
via freeCodeCamp, Apr 2026
The tools add structure; nobody goes back to simplify it.
Slide 13
Slide 13 text
02
s
e
l
m
x
e
e
s
c
o
e
r
h
SECTION TWO OF FIVE
The Pattern Gallery of Shame
t e f u d a p
Slide 14
Slide 14 text
E x a m p l e 1 · G E N ER AT E D BY A I
Pro t: "Set
We m t po
Factory for a Single Object
← One
is
lu
ba c
n oy
s
r l w e
ca s
ra s
t et .
e
s
a
p
d
b
o
s
e
n
l
r
i
n
t
r
a
o
d
a
f
o
r
s
e
c
t
e
e
l
a
e
m
r
s
a
u
s
r
r
s
e
l
e
n
h
p
t
a
o
b
d
h
e
t
r
a
i
r
n
p
h
s
r
y
u
a
p
s
u
s
e
y
s
a
s
d
c
a
c
y
e
h
l
g
o
p
i
a
t
c
m
r
A fa
Use
ro
s .
h o
mi
t
r e to .
ro l e ."
Slide 15
Slide 15 text
E x a m p l e 1 · G E N ER AT E D BY A I
Factory for a Single Object
d
o
t
e
n
r
r
m
i
e
t
t
a
s
a
e
r
l
s
c
b
e
l
t
e
a
s
e
u
r
r
y
e
n
a
h
s
h
o
d
t
d
a
s
r
a
p
s
u
a
r
p
c
u
a
s
y
r
o
e
h
t
g
s
p
i
c
m
A fa r l w e
ca s
h
on U
l e y si t .
Pro t: "Set
We m t po
mi
t
r e to .
ro l e ."
Slide 16
Slide 16 text
E x a m p l e 2 · G E N ER AT E D BY A I
t
i
l
t
i
m
e
i
e
h
t
s
r
i
r
t
s
s
e
r
l
e
s
u
r
o
s
e
w
t
p
o
e
r
e
l
p
i
d
x
y
l
Strategy for a Two-Branch Conditional
The m : "Let
s ot
te s
b at , ne t o d f
t. Ke p
f e b ."
Slide 17
Slide 17 text
← ad f
r
i
a
h
b
e
o
e
t
z
n
d
u
n
e
t
t
a
t
y
d
r
s
y
i
p
g
g
a
e
e
w
a
r
r
t
d
e
S
v
n
a
t
e
e
r
s
s
a
s
r
c
c
r
j
a
u
r
o
o
a
p
s
y
g
d
e
r
t
← Com
la
s
to IS he
ar
t , an
r
e vo
s at
be t
ne
s
pe
t
r
pa r .
im
Slide 18
Slide 18 text
t
r
w
e
t
s
.reversed() do
e
d
o
h
e
m
h
y
l
a
Re l t e
h
o k.
Slide 19
Slide 19 text
S I DE BY S I DE
What it wrote vs. what you needed
WHAT AI WROTE
✗ Comparator IS the Strategy pattern. It has
shipped with the JDK since 1998. The AI wrapped a
strategy in a Strategy.
✗ The abstraction adds four classes and a
registry, and zero behavior
✗ The strategies are never swapped at runtime;
the call site knew the direction all along.
WHAT YOU NEEDED
✓ One interface. Two classes. An enum. A
registry. Really the method .reversed()
does the work.
Slide 20
Slide 20 text
E X A M P L E 3 · G E N ER AT E D BY A I
Builder for Three Parameters
ul
on s
bu
, to
🤷
ob
c
il
es
se
ne
f u to
← Not g o t a , t e r
s al -bu
ca
The
p o p : "Ad a R r it
ti , da , an
n . Mak
ce
n a n i b ."
Com
an
e
d
o
l
i
d
n
r
i
e
e
h
p
a
r
t
m
o
o
g
i
y
e
l
n
h
o
t
c
e
l
j
b
a
o
a
l
t
k
n
i
i
n
o
a
e
e
d
h
a
e
u
y
m
w
w
i
o
e
r
l
h
t
e
n
r
h
v
e
t
o
a
l
r
p
n
a
e
n
l
w
a
a
o
n
o
i
t
l
t
i
s
o
l
p
i
t
d
i
i
i
d
d
s
f
e
t
i
m
e
n
g
e
d
n
h
h
t
f
i
s
t
h
t
a
a
r
i
e
l
x
e
e
i
r
n
e
p
d
m
e
l
a
i
d
n
t
r
l
Bu l
s i , fa t u t
r
to
sin
t a s
ye
r
Slide 21
Slide 21 text
E X A M P L E 3 · G E N ER AT E D BY A I
Builder for Three Parameters
The m : "Ad a R r
it
da , an
n . Mak
le
ma in ."
t
l
a
u
l
i
t
n
a
i
a
t
e
h
c
n
u
a
l
w
t
e
t
c
s
n
t
o
i
o
p
e
c
e
u
o
y
y
t
i
d
i
r
e
b
v
i
e
l
w
g
t
t
o
b
u
p
a
d
r
o
o
m
r
a
i
c
t
p
e
n
d
i
t
A re d es
an m a l
d
t e,
on r on, eq i y,
in
in .
Slide 22
Slide 22 text
03
When Patterns Actually
Help + A Red-Flag Checklist
A qu d a s
pa r de
to
l
d
a
r
e
n
I
n
i
g
m
e
r
o
b
y
p
p
a
c
i
t
o
n
g
o
i
c
n
k
c
i
e
t
t
SECTION THREE OF FIVE
ef
er g A -ge
te
Slide 23
Slide 23 text
As k you r s e l f
Quick diagnostic to apply before merging AI-generated pattern code
1
Is there more than one implementation today, or a realistic reason to
expect one?
2
Does removing the pattern eliminate complexity without
eliminating capability?
3
Would a new dev need to look this pattern up to understand it?
4
Did the AI add the pattern without being asked?
Slide 24
Slide 24 text
T H E P L OT T WI S T
"Wa t — t i t
he
St a g t e n
ac l ig ."
It's not always wrong.
When there are genuinely many variants
t e k s el
t
if
n e.
Know when it is just "Speculative Generality"
happening.
c
e
r
e
f
d
e
h
t
s
i
e
g
n
m
r
i
i
l
t
t
s
t
h
a
h
p
i
r
y
l
l
y
i
e
l
t
a
s
i
u
r
t
h
image from shutterstock
Slide 25
Slide 25 text
No content
Slide 26
Slide 26 text
04
Prompting for Simplicity
Th e r ic
en er .
r
e
v
m
o
f
y
a
I
r
e
e
t
t
s
p
m
o
p
l
a
t
g
c
a
n
i
p
e
e
n
i
r
g
SECTION FOUR OF FIVE
t os
A aw
r
o r-
Slide 27
Slide 27 text
T H E R EC OV ERY P L A N
How to prompt for restraint
during · Split
Before · CONSTRAIN
→
"Simplest thing that works.
No new abstractions."
Get a patternless
baseline first, then make
the abstraction call a
separate step
→
y
f
i
p
n
o
m
i
t
o
c
x
t
e
n
e
a
t
t
o
y
o
d
t
u
n
a
f
i
t
d
e
e
v
t
n
y
i
g
c
n
l
i
m
o
i
c
s
t
e
?
l
a
t
w
u
s
Wha d co n o
p
e
h
o
n
,
l
e
h
pi t
Mak
After · Interrogate
"Why is there a Factory
here? What problem does
it solve?"
Slide 28
Slide 28 text
The Best Pattern Is the
One You Can Delete
05
Pat
n
s
e
u
r
t
y
r
a
b
a
c
o
r
a
s
r
e
t
SECTION FIVE OF FIVE
ev
ul
, no
l .
Slide 29
Slide 29 text
THANKS, NEBRASKA.CODE()!
Go write less code .
1 · A pattern is a tool, not a reflex.
Q&A
2 · Make the AI justify the abstraction.
w a 's e w es
pa r u've
ca t A a n ?
3 · The best pattern is often no pattern.
t
g
d
i
r
i
d
e
d
o
I
h
y
t
n
e
h
t
t
g
t
u
h
@[email protected] · references: christine-seeman.com/ai-patterns