2026年大创意：第一部分

Big
Ideas
2026:
Part
1
2026
年⼤创意：第⼀部分
The
biggest
problems
builders
will
tackle
in
2026,
according
to
our
partners
根据我们的合作伙伴的观点，2026
年建设者将解决的最⼤问题
Our
job
as
investors
is
to
immerse
ourselves
in
the
ins-and-outs
of
every
corner
of
the
tech
industry
in
order
to
understand
where
things
are
moving
next.
So
every
December,
we
ask
our
investing
teams
to
share
one
big
idea
they
think
tech
builders
will
tackle
in
the
year
to
come.
作为投资者，我们的⼯作是深⼊了解科技⾏业各个领域的细节，以预测未来的发展⽅向。因此，每年
12
⽉，我们都会要求投资团队分享他们认为科技建设者在来年将要解决的⼀个重⼤想法。
Today,
we’re
sharing
ideas
from
our
Infrastructure,
Growth,
Bio
+
Health,
and
Speedrun
teams.
Stay
tuned
for
takes
from
our
other
teams
tomorrow.
今天，我们分享来⾃基础设施、增⻓、⽣物+健康和快速迭代团队的想法。敬请期待我们其他团队明天的观点。

已订阅
Infrastructure

基础设施
Jennifer
Li:
Startups
tame
the
chaos
of
multimodal
data
詹妮弗·李：初创公司驯服多模态数据的混乱
Unstructured,
multimodal
data
has
been
enterprises’
biggest
bottleneck
and
their
biggest
untapped
treasure.
Every
company
is
drowning
in
PDFs,
screenshots,
videos,
logs,
emails,
and
semi-structured
sludge.
Models
keep
getting
smarter
but
the
inputs
keep
getting
messier,
which
causes
RAG
systems
to
hallucinate,
agents
to
break
in
subtle,
expensive
ways,
and
critical
workflows
to
still
heavily
rely
on
human
QA.
The
limiting
factor
for
AI
companies
is
now
data
entropy:
the
steady
decay
of
freshness,
structured,
and
truth
inside
the
unstructured
universe
where
80%
of
corporate
knowledge
now
lives.
⾮结构化的多模态数据⼀直是企业最⼤的瓶颈，也是最⼤的未开发宝藏。每家公司都被
PDF、截图、视频、⽇志、电⼦邮件和半结构化的数据淹没。模型变得越来越聪明，但输⼊数据却越来越混乱，这导致
RAG
系统产⽣幻觉、代理以微妙⽽昂贵的⽅式崩溃，关键⼯作流仍然严重依赖⼈⼯质量保证。对
AI
公司来说，现在的限制因素是数据熵：在这个
80%的企业知识所在的⾮结构化世界中，数据新鲜度、结构性和真实性的持续衰退。 That’s
why
untangling
unstructured
data
becomes
a
generational
opportunity.
Enterprises
need
a
continuous
way
to
clean,
structure,
validate
and
govern
their
multimodal
data
so
downstream
AI
workloads
actually
work.
The
use
cases
are
everywhere:
contract
analysis,
onboarding
flows,
claims
handling,
compliance,
support,
procurement,
engineering
search,
sales
enablement,
analytics
pipelines,
and
every
agent
workflow
that
depends
on
reliable
context.
Startups
that
build
the
platform
that
extracts
structure
from
documents,
images,
and
videos;
reconciles
conflicts;
repairs
pipelines;
or
keeps
data
fresh
and
retrievable
hold
the
key
to
the
kingdom
of
enterprise
knowledge
and
process.
这就是为什么理顺⾮结构化数据成为⼀个时代性的机遇。企业需要⼀种持续的⽅式来清理、结构化、验证和治理其多模态数据，使下游
AI
⼯作负载真正发挥作⽤。应⽤场景⽆处不在：合同分析、员⼯⼊职流程、理赔处理、合规性审查、客⼾⽀持、采购、⼯程搜索、销售赋能、分析管道，以及每⼀个依赖可靠上下⽂的代理⼯作流。那些构建能从⽂档、图像和视频中提取结构的平台、调和冲突、修复管道或保持数据新鲜可检索的初创公司，掌握着企业知识和流程王国的钥匙。 Joel
de
la
Garza:
AI
revives
cybersecurity
hiring
乔尔·德拉·加尔萨：AI
复兴⽹络安全招聘
For
the
better
part
of
the
last
decade,
the
biggest
challenge
CISOs
faced
was
hiring.
From
2013
to
2021,
the
number
of
unfilled
cybersecurity
jobs
grew
from
under
1M
to
3M.
That’s
because
security
teams
hire
highly
skilled
technicians
to
spend
their
days
doing
soul-crushing
level
1
security
work
like
reviewing
logs,
and
no
one
wants
to
do
this
work.
The
problem
is
that
cybersecurity
teams
create
that
drudgery
by
buying
products
that
detect
everything,
which
means
that
their
teams
need
to
review
everything—which,
in
turn,
creates
false
labor
scarcity.
It’s
a
vicious
cycle.
在过去的⼗年⾥，⾸席信息安全官（CISOs）⾯临的最⼤挑战是招聘。从
2013
年到
2021
年，未填补的⽹络安全职位数量从不到
100
万增⻓到
300
万。这是因为安全团队雇⽤⾼技能的技术⼈员来做令⼈厌倦的⼀级安全⼯作，⽐如审查⽇志，⽽没有⼈想做这种⼯作。问题在于⽹络安全团队通过购买能检测所有内容的产品来造成这种苦差，这意味着他们的团队需要审查所有内容⸺进⽽造成了虚假的劳动⼒短缺。这是⼀个恶性循环。 In
2026,
AI
will
break
this
cycle
and
close
this
hiring
gap
by
automating
much
of
this
repetitive
and
redundant
work
for
cybersecurity
teams.
Anyone
who
has
ever
spent
time
working
on
a
large
security
team
knows
that
half
the
work
is
easily
solved
with
automation,
but
it’s
impossible
to
figure
out
what
to
automate
when
you’re
drowning
in
work.
AI-native
tools
that
figure
this
out
for
security
teams
will
finally
free
them
up
to
do
what
they
want
to
do:
chase
down
bad
guys,
build
new
systems,
and
fix
vulnerabilities.
到
2026
年，⼈⼯智能将通过⾃动化⽹络安全团队的⼤量重复性和冗余⼯作来打破这个循环并缩⼩招聘差距。任何曾在⼤型安全团队⼯作过的⼈都知道，⼀半的⼯作可以很容易地通过⾃动化解决，但当你被⼯作淹没时，很难搞清楚要⾃动化什么。为安全团队解决这个问题的原⽣
AI
⼯具最终将使他们能够做他们想做的事情：追捕坏⼈、构建新系统和修复漏洞。 Malika
Aubakirova:
Agent-native
infrastructure
becomes
table
stakes
Malika
Aubakirova：智能体原⽣基础设施成为基本要素
In
2026,
the
biggest
infrastructure
shock
won’t
come
from
outside
companies,
but
from
within.
We’re
shifting
from
human-speed
traffic
that’s
predictable
and
low
concurrency
to
“agent-speed”
workloads
that’re
recursive,
bursty,
and
massive.
到
2026
年，最⼤的基础设施冲击不会来⾃外部公司，⽽是来⾃内部。我们正在从⼈类速度的可预测、低并发流量转变为"智能体速度"的⼯作负载，这些⼯作负载是递归的、突发的和⼤规模的。
The
enterprise
backend
of
today
was
built
for
a
1:1
ratio
of
human
action-to-system
response.
It’s
not
architected
for
a
single
agentic
“goal”
to
trigger
a
recursive
fan-out
of
5,000
sub-tasks,
database
queries,
and
internal
API
calls
in
under
milliseconds.
When
an
agent
attempts
to
refactor
a
codebase
or
remediate
a
security
log,
it
doesn’t
look
like
a
user.
To
a
legacy
database
or
rate-limiter,
it
looks
like
a
DDoS
attack.
当今的企业后端系统是为
1:1
的⼈类⾏为与系统响应⽐率⽽设计的。它的架构⽆法应对