CHAPTERS
- 0:09 – 1:10
What AI-native services companies are (and why they’re huge)
The speaker defines AI-native services companies as outcome-delivery businesses where AI does most of the work, enabling entire service industries to be rebuilt from scratch. He contrasts this with selling copilots and frames the opportunity as newly unlocked by recent model advances across trillion-dollar markets.
- •AI-native services deliver outcomes, not software tools customers operate
- •New model capabilities now make this business type viable
- •Targets massive legacy services markets (tax, insurance, law, healthcare, etc.)
- •These companies will look and operate differently from typical software startups
- 1:10 – 2:41
Market selection: four traits that make AI services scalable
Market selection follows classic startup advice (pick something you’ll do for a decade), but AI services markets have special characteristics that determine scalability and defensibility. The talk outlines four traits that strongly correlate with success.
- •Low trust: work is already outsourced; customers care about outputs, not process
- •Low judgment at the task level: most steps should be automatable
- •High intelligence threshold: hard problems where models+humans beat incumbents
- •Regulation can be a moat by raising accountability and expectations
- 2:41 – 3:11
Examples of strong markets YC likes (and why regulation can help)
The speaker lists specific verticals that fit the AI-native services pattern and gives an example of a YC company using AI plus expert consultants in a regulated setting. He encourages founders to look beyond the obvious, popular market picks.
- •YC ‘known good fits’: tax, audit, insurance, mortgages, parts of healthcare/logistics
- •Many underexplored service markets remain untouched
- •Panacea example: FDA regulatory services powered by AI + experienced consultants
- •Regulation can increase defensibility and raise the quality bar
- 3:11 – 4:11
The Sam Altman test: will better models strengthen you or commoditize you?
Founders should evaluate whether ongoing model improvements amplify their advantage or erase it. The chapter also flags business types where physical operations undermine software-like leverage and urges honesty about when humans are covering product gaps.
- •Ask: as models improve, does your service get stronger or get commoditized?
- •Be cautious with equipment-heavy/on-site labor businesses (harder to get leverage)
- •Don’t confuse necessary judgment with patching product shortcomings
- •Humans-in-the-loop can be great—but must be intentional
- 4:11 – 5:12
Founding team: the three fluencies that matter
AI services companies require a team that combines credibility in the domain, deep understanding of frontier models, and operational excellence. The speaker explains why each is crucial to winning skeptical buyers and building a scalable operation.
- •Domain fluency builds credibility with skeptical/regulatory buyers
- •Model fluency: design to ‘ride the curve’ of improving capabilities
- •Operational rigor is essential (variance control, throughput, SOPs)
- •Founders should ideally work with trusted collaborators, not strangers
- 5:12 – 5:42
Operating like a product: General Legal as an AI-native law firm example
A concrete example illustrates what a strong AI services founding team and operating mindset look like. The company blends legal expertise with technical leadership and designs staffing/shift work to improve cycle time and scalability.
- •General Legal combines top law firm experience with AI/tech leadership
- •Throughput and staffing strategy are core to product and customer experience
- •Shift work reduces cycle time and can improve talent attraction/retention
- •Operational design becomes a competitive advantage
- 5:42 – 6:13
Building the product: the human is the interface
Unlike software startups, the customer experience is mediated by humans, while the product exists to scale those humans non-linearly. This flips product thinking toward operations, bottlenecks, and measurable flow efficiency.
- •In AI services, humans face customers; software amplifies their output
- •Apply an operations mindset: build for bottlenecks
- •Throughput and cycle time become first-class product metrics
- •Automation of the process ultimately is the product
- 6:13 – 6:43
Variance control and non-linear scaling as existential requirements
The speaker argues that inconsistency in outputs is the fastest path to churn because it destroys trust. He also emphasizes the necessity of non-linear human scaling and designing software that internal operators actually like using.
- •Variance (inconsistent output) kills trust faster than price or speed issues
- •Humans-in-the-loop must scale non-linearly vs. revenue growth
- •Operators are your real users—UX matters for them
- •‘Do things that don’t scale’ early, but plan for real scalability
- 6:43 – 7:13
Avoiding the early demand trap in pilots and delivery
Early demand can overwhelm delivery capacity and freeze product development, forcing founders into a labor-heavy corner. The guidance is to cap pilots, treat pilots as learning vehicles, and iterate quickly to find real leverage points.
- •Too many early pilots can overwhelm delivery and block scaling work
- •Cap pilots to a small handful; resist signing everyone
- •Sell outcomes; the pilot is effectively the product
- •Use pilots to discover where AI provides unique leverage vs. obvious automation
- 7:13 – 8:14
Pricing AI services: value-based structures and traps to avoid
Pricing differs from SaaS because the competitive benchmark is labor cost (internal or outsourced), not other software. The speaker outlines viable pricing models and warns against approaches that permanently limit upside or signal low quality.
- •Compete against labor economics, not seat/token pricing
- •Per-unit pricing (per return/claim/loan) is clean and explainable
- •Outcome-based pricing aligns incentives but complicates forecasting
- •Avoid cost-plus pricing and simplistic undercutting; price on value
- 8:14 – 8:44
P&L fundamentals: where AI services companies live or die
The talk walks through a services-oriented P&L and highlights what investors and operators will scrutinize early. It frames delivery consistency and margin trajectory as central to whether the model works.
- •P&L structure: revenue → gross profit (minus COGS) → operating income (minus OpEx)
- •Revenue is achievable; repeatable delivery is the real test
- •Expect early revenue lumpiness; process and product smooth it over time
- •You’ll be evaluated on operating income sooner than typical software startups
- 8:44 – 10:15
COGS ownership, margin discipline, and AI operating leverage
COGS is the core battleground: model costs, hosting, and human labor must be measured and owned. The central bet is that as the product matures, COGS declines and margins expand—creating ‘AI operating leverage.’
- •COGS components: model costs, hosting costs, humans-in-the-loop
- •Assign owners and track trend lines for each cost component
- •Beware zero/negative margin pilots becoming a habit
- •AI operating leverage: product maturity reduces COGS and raises gross margin
- 10:15 – 11:21
Don’t buy your way in: why acquisition-first strategies usually fail
Founders are tempted to buy a legacy services firm to shortcut revenue, then layer AI on top. The speaker argues this rarely works because you can’t acquire product-market fit and legacy constraints resist transformation, with limited exceptions for fast regulatory moats.
- •Buying an existing firm to ‘add AI’ is usually a trap
- •One exception: quickly obtaining regulatory moats (e.g., insurance licensing)
- •Legacy businesses carry mismatched expectations on metrics, hiring, performance
- •Building from scratch is typically superior to acquiring
