Slides: "The engineer of the future is the person who is able to choose what is worth doing." — Addy Osmani
Source Video
"The engineer of the future is the person who is able to choose what is worth doing." — Addy Osmani
Relationship To World's Fair 2026
These slides are extracted from a verified official AI Engineer World's Fair San Francisco 2026 recording. Use them as slide and OCR evidence; official schedule pages remain canonical for titles, times, rooms, tracks, speakers, and affiliations.
Related Scheduled Sessions
- 2026 06 30 addy osmani closing keynote — Closing Keynote
Extracted Slides
OCR text:
LAB & PLATINUM SPONSORS
Amazon AGI Lab ANTHROP\C Google DeepMind MINIMAX
OpenAl ZB ss akamai arize 2YS Braintrust brightdata
B Browserbase = docker -:NEO4j) ORACLE PayPal ¢&&
‘yreducto @Q sonar’: © togetherai ({f Unblocked €} WorkOS
OCR text:
| VVONQSFrall | IVIICTFOSOTL
AlEngineer —————
oer aTay World's Fair
AlEngineer — | ; 7
World's ar a 7 Akamai
oe | | —— AlEnginccr —_
OCR text:
| The engineer of the future will choose
Bee aL is worth doing, then
for
work increasingly automated by agents.
riastair i
penAl Wo One Meant
ae el
Addy Osmani Google
OCR text:
7 ee | =
Mela kd 1 y ~4. iP ae
d We A
| foi be <=
m Microsoft ot ad a :
ee i% uf —s \ roe ea oat
ane — PS a
PERSE ate @ a Oe a Judge F
a1 Bugs aa y,
va Wena a
re Fe a, Keynote ee
Me Ae Cal Google
OCR text:
od Quality = the system of checks that
produces evidence.
| Verdict = the human/accountable
a acerrol decision made from that evidence.
Answerability = the ability to explain
eal “and stand behind the verdict later.
aay re ee
ae r Keynote
7 pene Osmani ; Google
OCR text:
a Boris Cherny @ Or
Are @dcherny
D H As engineering, product, design, DS, etc. melt Into a new kind of role, |
World s Fair was reflecting on what roles might look like in the future. For example,
when | look at the Claude Code team | see what I think is five archetypes:
1. Prototyper: comes up with brand new ideas; chums out many ideas,
most of which don't ship
F 2. Bullder: quickly turns a prototype/idea into production-grade
product/infra
i 5 3. Sweeper: cleans up the Ul, simplifies the code and system, unships,
T | M Icros revat optimizes performance
4. Grower: takes a product that has been built and iterates on it to
improve Product-Market Fit
5. Maintainer: owns a mature system to make it secure, reliable, fast,
and efficient as it scales
Many people span across 2 roles, and sometimes 3 roles. | also notice
tsrair Ivit that these roles are not really tied to job function -- eg. across Anthropic,
some designers match category 1, some 2, some 3; same for engineers,
: PM, OS.
a
eral Worlc @addyosmani
I's Fa Ak eynote
ee >) Ori ra) Google
OCR text:
er ae ;
Roles are unbundling from craft
and rebundling around ownership.
mu Microsoft Prototype. Build. ; Maintain.
The title matters less than
ory « the part of the system you can own.
enAl | Worl @addyosmani
J's ' FS
_, Addy Osmani Google
OCR text:
PNET erg
World's Fair HARNESS ENGINEERING
afore scaffolding that turns a model into an agent,
| an " a) Ee re
a Microsoft — “
aie ol ee ;
forssr eds a : A
lasrair M7
voray ay et
eee , Keynote .
Re se Rear val Google
OCR text:
LU Seed
eeu AGENTIC SOFTWARE FACTORY
monitor eet Le 3
product intent
ae de eapyeeng Coat falas
incidents prod
a Microsoft ree
RYT r-rel Lola 9 ae 7
human verdict
1aS rair NV
TAN ren The win ix moving human judgment to the highest leverage checkpoint.
rit 3 F Lec
_ Addy Osmani Google
OCR text:
EMTs ag
World's Fair | p
Reviewers are already overloaded
g oro mathe Late (aren tote Rested
PPE ikecesxene
eR Ree Cote eA kols aeer co flcemel aad iter?
a re,
Say reviewing Al code takes longer than human code
aSsrair M7
The risk is not that engineers distrust Al. It is that they distrust it and
A CH Reem Ee ass ane
enAl Wor
mG P Keynote
Rees Aer tral] Google
OCR text:
—_—
Male aCekd alls lJ
+ '
J
las rair UEnCeae y . aa
denAl oy
Id's i N Keynote
- Ms AOA] Cre
OCR text:
World's Fair | yo
The agent can ship more than you can’ *""":
etl
So whatare you still for? ;
wiads Fair AllldZ |
ipenAl Ae Renee
Per Mo
; Addy Osmani Google
OCR text:
ad Alphais the gap.
Alpha:
what makes you meaningfully better
than what current models can do.
NE 1 ETS ata cTaniers (-\cXersl canoe
eer Ay Wo ORC SuEn]
in n Ne ;
. Addy Osmani Google
OCR text:
renee al ;
BF Caa a= before the
metric exists.
Hard to create. Easy to copy.
Valuable because it decides what
Tes ae everyone copies next.
pT Tar AN Wor @addyosmani
ee E es)
_ _ Addy Osmani Google
OCR text:
World's Fair
Speed
Recall
Verification
Eventually? Maybe?
Taste
Judgment
nasralr cere E
penAl Wo: ORC Chari
Ad it dina ee a
PVs ohVOrs ara) Coe 2
OCR text:
World's Fair
Stop asking what Al can't do.
Ask what only a human can be
answerable for.
srair PUeRroP ern)
Nae ry ae Sere TN Te
ae Vlic Keynote
7 eo Rea) Google
OCR text:
World's Fair |
Everyone is a developer now.
That doesnt make them
an engineer.
What should engineers avoid to stay
ter Vee ale ?
id leBce relia AllidZd
penA Wo Once heart
7 F | Keynote
Addy Osmani Google
OCR text:
World's Fair
7 Mt P Bbat-¥-le (sn kee esos
an hO8O Only people inherit consequences.
aqastalr 7
wi EE ver
ro ois
Addy Osmani Google
OCR text:
feta te Kod are g
Accountability is not what remains
: after agents get good.
PE iacesronie
It is what lets the rest scale.
oe arel is 7. ae
moe aE Be os eS
aa ar Keynote
Addy Osmani Google
OCR text:
hha The half life of an edge is a release.
The haff life of a signature is a career.
a Microsoft —
asralr £. speed Verification taste
cn vate
aig c Keynote
Me RO ie a] Google
OCR text:
we Meares (ols aesVAR-Te lee ate aban dees aie ok ere ui akg
tee ad
World's lle WOST AGENCY
ge are) eee re 7 DISCERN
are eee ee eee 3 RESOLVE
7 ; Be ree ren ea ae iS RECOMMEND
Paka cestenns
ee ee cy Lael aek1
Cee Ee i} DIAGNOSE
at en v1 EXECUTE
asralr 7
Tepeng ga yar te ene 5 FLAG
“ ae (a) a rere Cerra LEAST AGENCY
ekoas c Keynote
Addy Osmani Google
OCR text:
Ea Syed
Welaleka ale THE LOOP BOUNDARY IS EVIDENCE
Agents run capability loops. Engincurs own agency loops.
AGENT INNER LOOP « ENGINEER OUTER LOOP
er DECIDE ee
7D 1d, re a _
‘sen gn ve Basle o mye oswslh VERIFY Bs won mmr me en
Teele z ey ee ee eee eae
The inner loop is capability. The outer loop ts accountability.
Dee te fete
Rig Di lid
Addy Osmani Google
OCR text:
World's Fair
Every time we made it easier to write
software, we ended up wniting
exponentially more of it.
RCM tails yi The future belongs to engineers who make agent work legible, verifiable, and worth shipping.
ore Worl CRC SaEL
ds! fs ia ch -
Addy Osmani Google
OCR text:
Ce
re Steer erent Ped omea reerent Fey co rent red fered renent red y aN preentes ed Renee mee
De tT ES Sa endo} Rata aes Peete Words Fae hr Reeet eed “ret poet res nga need
Pred Net ed Ac ids Ree ed ak Sed Workrs Far isdeaerasaete DTT ee or ter fore Prana e a
o Scr eey aa Word ar TS ‘ om petted oa roe]
~ rere be) Settee ee i Eeeerna red Seite feo
Seromen retired years 25 ae 7 ’) i ra Wetter rao
, . i .
norte Veet) oe j cern e rnd
i
_ acne Reed al i \ Deter Centr La
Peer red os nous tees thet erie ce : a eked wot Skea Pemtee sy Peake ee na Ped
Ps : freer trey Ee eres Ans) vanstutae [TOR OF GOOF g. Peery ey cea Paever en pean re
Pass reeet rey . se tte Me tard sae hy tee cm Ce oe rok Wewts tier wed Cal
a Nees) cee er ee dace cere ey PONE STAT TCM owt Tea e ° g aoe fee tT Pe coed romney aan poe :
o eerie veer ene Te tn ert red De] | eer ~ ited Moa ited By sew ond
cae a ae ahead a es E Dade - Taine Leaded ed ite e *
co rte Pty ee P smentavens FR , pire area ete co Pon ree)
von fen en an LvEet Sapam Denner Set iedd Pea feet Te! " ry
FB ane oer oe eerie shee an te a ppaeet rr a pen
ry or Lara T td Ci ege-r ci a see0 eseee sei tod
OCR text:
1t) yvorias ralr | IVIICFOSC
© Opens | wort
iv - | | Akamai
5. a Lee ee
OCR text:
The engineer of the future will choose
whatis worth doing, then
for
work increasingly automated by agents.
Tara ent @addyosmani
rea
OCR text:
: ae | =
World's Fair y wip iP , a
d e ; zs
_ ° |
fo be -
a Microsoft ; —_—" a ad Tae aiaaa
2% > OS
ee a4 SS 4 an od aan
las Fair Vv ee . Judge oe
a Bugs ag
eter N van ys
ROS FL ee |
Me Ae Google
OCR text:
od Quality = the system of checks that
produces evidence.
— Verdict = the human/accountable
eM ecerrol decision made from that evidence.
Answerability = the ability to explain
al “and stand behind the verdict later.
Tr vet a
m Fi a Lol)
. 7 pore Osmani Google
OCR text:
a Boris Cherny @ Ow
NY? @bcherny
D H As engineering, product, design, DS, etc. melt Into a new kind of role, |
World Ss il was reflecting on what roles might look like in the future. For example,
when | look at the Claude Code team | see what I think is five archetypes:
1. Prototyper: comes up with brand new ideas; chums out many ideas,
most of which don't ship
2. Bullder: quickly turns a prototype/idea into production-grade
product/infra
i A 3. Sweeper: cleans up the UI, simplifies the code and system, unships,
T | | Icros revue optimizes performance
4. Grower: takes a product that has been built and iterates on it to
improve Product-Market Fit
5. Maintainer: owns a mature system to make it secure, reliable, fast,
and efficient as it scales
Many people span across 2 roles, and sometimes 3 roles. | also notice
asrair al that these roles are not really tied to job function -- eg. across Anthropic,
some designers match category 1, some 2, some 3; same for engineers,
, PM, OS.
TaN Worl: ORME]
_ L
naa Ai eynote
es \2O Ta) Google
OCR text:
ere ae ;
Roles are unbundling from craft
and rebundling around ownership.
a Microsoft Prototype. Build. ; Maintain.
BPR eyare ceed oe alla)
ror » the part of the system you can own.
ole ay Tel OnE Ea et
ra , Keynote
. _ Addy Osmani ae Google
OCR text:
EU tye ad
kas HARNESS ENGINEERING
The searfoiding that turn. amodelinto aun agent.
. 7 ~ | aetaea
a Microsoft i ' _
Eee A REELS! - 7
nasrair u
penAl iY,¥ a)
rd? Keynote
Addy Osmani Google
OCR text:
EU eed
eeu AGENTIC SOFTWARE FACTORY
monitor eet Le 3
product intent
ae de eapyeeng Coat falas
incidents prod
a Microsoft pee
RYT r-rel Lola 9 ae 7
human verdict
1aStralir hy
ey re ee eT een ree eT eer eee
dd'sF Fe
Me 0e rela) Google
Slide-Derived Subjects To Review
Subject extraction uses video title, related session titles/descriptions, transcript context, and OCR text when available. OCR is best-effort and should be reviewed against the embedded slide images.