Y CombinatorHow Enterprise AI Skeptics Hand Startups the Market
Enterprise engineers who disbelieve AI give startups time to ship deep integrations; Reducto shows this beats SaaS plug-and-play on every enterprise deal.
CHAPTERS
- 0:00 – 0:35
Enterprises can’t build AI products when engineers don’t believe in AI
Harj argues that many enterprise engineering teams are culturally skeptical of AI, which becomes self-fulfilling: they don’t adopt the tools, so they can’t ship working AI products. This creates an opening where startups that can actually make AI work become the only viable option for buyers.
- •Internal skepticism toward AI blocks experimentation and adoption
- •Belief that AI is “overhyped” becomes a convenient narrative inside large orgs
- •Inability to build internally forces enterprises to look externally
- •Startups get opportunities they previously wouldn’t in enterprise accounts
- 0:35 – 1:54
The “95% failure” narrative: what the MIT study is (and isn’t) saying
Jared pushes back on influencer takes that cite a viral MIT statistic as proof AI is a scam. The discussion reframes the study as consistent with real-world experience: AI can work, but implementation is hard and requires the right approach.
- •Viral summaries of research can be misleading
- •The study’s conclusions align with practical constraints of deploying agents
- •Failure often reflects execution issues, not inherent uselessness
- •The right categories/approaches show better outcomes in practice
- 1:54 – 2:40
Why AI go-to-market isn’t classic SaaS: deep integration into systems of record
Diana explains that enterprise AI success often demands founders embed into workflows and integrate deeply with internal systems, unlike plug-and-play SaaS. The reward is large once integrated, but timelines and effort are materially higher.
- •AI deployments require understanding real business processes end-to-end
- •Deep integrations with systems of record are decisive
- •More services-like involvement early (“do things that don’t scale”)
- •Big upside once entrenched, but slower path than traditional SaaS
- 2:40 – 4:29
The enterprise adoption gap: internal IT + consultants produce “bad software”
The panel outlines a common enterprise pattern: attempts start with internal IT, then escalate to large consultancies, often yielding poor results. Jared uses Apple’s flawed Calendar app as an analogy that even well-resourced orgs ship mediocre software—so it’s unsurprising enterprises struggle even more.
- •Internal enterprise IT systems are often low-quality
- •Consulting-led builds add coordination overhead without strong product execution
- •Even elite companies ship buggy, unsatisfying software
- •The default outcome is a committee-designed “camel,” not a great product
- 4:29 – 5:36
Why implementation fails: politics, turf wars, and legacy silos
Harj highlights organizational complexity as a core blocker: multiple teams must align, causing political friction and competing requirements. Consultants can mediate and produce specs, but real success still requires technical capability to implement across old, siloed systems.
- •Cross-team adoption forces alignment across competing stakeholders
- •Consultants help coordination/specs but often can’t build the final system well
- •Legacy infrastructure and data silos raise integration difficulty
- •The “horse designed by committee” dynamic undermines product quality
- 5:36 – 7:42
Case studies in banking: Tactile and Greenlight outperform internal builds
Examples show startups delivering in months what banks attempt over years at high cost. Greenlight’s story illustrates how a bank trusted an existing consultancy, lost a year on a failed build, then returned to the startup for a working deployment.
- •Tactile: real-time decisioning API delivered faster/cheaper than bank internal efforts
- •Banks spend 3–5 years and tens of millions building similar systems
- •Greenlight: bank chose Ernst & Young first; build failed; startup later succeeded
- •MIT study detail: vendor-bought solutions had higher success than in-house/consultant builds
- 7:42 – 9:37
What wins: rare polymath teams + product taste + domain empathy
Jared and Diana argue that successful AI products need builders who combine engineering skill with product sense and deep understanding of human workflows. This gap creates a “startup-shaped hole” across many enterprise processes that should exist but don’t yet.
- •Few people combine product, engineering, and domain understanding
- •Domain experts often can’t ship software; engineers often lack workflow empathy
- •Some elite users can build tools themselves, but most organizations can’t
- •Startups fill the gap by shipping end-to-end systems that work
- 9:37 – 11:22
Incumbents “slap AI on top”: Castle.AI bake-offs and AI-native advantage
Diana describes how legacy vendors and incumbents respond by bolting AI onto old products, which performs poorly in head-to-head evaluations. Castle.AI wins by being AI-native and delivering better outcomes despite entrenched trust in long-standing vendors.
- •Banks default to incumbents due to trust and familiarity
- •Bake-offs reveal incumbents often deliver “AI slapped on top”
- •AI-native design and better product execution drive competitive wins
- •Even fast-growing startups can close large banks quickly with superior results
- 11:22 – 13:02
Reducto’s enterprise win: beating internal teams with execution and politics
Reducto lands a major FAANG customer shortly after YC by outperforming internal OCR/document-processing efforts. The story underscores that technical excellence must be paired with navigating internal politics and competing with internal teams.
- •Reducto: document processing for AI; won after internal solutions failed
- •Customer tried open source, AWS Tesseract, and other OCR approaches
- •Winning required finesse amid internal competition and politics
- •Outcome: large deal and sustained production deployment
- 13:02 – 14:39
How to win enterprise champions: friendships, authenticity, and founder energy
The panel emphasizes “do things that don’t scale” to earn trust—especially building strong relationships with a champion. They note an archetype of employee who dreams of startup life and will advocate for a startup vicariously; founders should also avoid over-corporate posturing and stay authentic.
- •Win by deeply supporting the internal champion and building real rapport
- •Some champions want to feel part of a startup journey and push harder for you
- •Founders’ ambition/optimism can be contagious inside large orgs
- •Avoid performative enterprise formalism; authenticity can be an asset
- 14:39 – 15:24
Shortcut into big companies: champions who were acquired can guide procurement
Harj shares tactics from Triplebyte: leverage founders who sold companies into a target enterprise to navigate procurement and politics. Such insiders can provide a playbook and unlock pilots that are otherwise hard to obtain.
- •Acquired founders/employees can become powerful internal sponsors
- •They can demystify procurement and internal stakeholders
- •Examples: access to Apple via acquired YC founders; Oracle pilot via insider help
- •Silicon Valley’s “pay it forward” network effect boosts enterprise selling
- 15:24 – 19:40
Why now is unusually good for startups: enterprise demand + internal AI skepticism
Garry and Harj argue the optimistic takeaway from the study is strong enterprise demand and increased willingness to bet on startups. Enterprises may prefer established vendors, but if those vendors can’t build AI-native products—and internal teams resist AI—startups become the only path to working solutions.
- •Enterprise appetite for AI is high despite poor implementation rates
- •It’s easier now to sell AI agents into large companies than in prior eras
- •Established vendors often can’t deliver AI-native products fast enough
- •Internal skepticism toward AI tools slows enterprise execution, creating startup opportunity
- 19:40 – 21:43
Moats in enterprise AI: training and switching costs + the ‘top 5%’ execution edge
A cited buyer quote highlights a practical moat: once an enterprise trains a system, switching becomes prohibitive. The closing frames the 5% success rate as encouraging for exceptional teams—founders who combine technical excellence and real user understanding can be in the small set that delivers working deployments.
- •Training/integration time creates high switching costs (a defensible moat)
- •“Wrappers” can become sticky when embedded in workflows and data
- •Success concentrates among teams with strong execution and product sense
- •Takeaway: AI is hard, but that difficulty is exactly the opportunity for great startups
