ExpertsFMChristina Cacioppo - Building an Enduring Software Company in the Age of AI - [Invest Like the Best, EP.496]
Invest Like the Best with Patrick O'Shaughnessy · October 20, 2026
5 Selected Lessons:
heuristic · 2 framework · 2 case · 1
Integration data moat ('trust graph')
Source
Invest Like the Best with Patrick O'Shaughnessy — Christina Cacioppo - Building an Enduring Software Company in the Age of AI - [Invest Like the Best, EP.496] (October 20, 2026) · at 3:08 ▸
When this applies
when building a platform business where aggregated third-party data could power future AI/automation features
The takeaway
Invest early in aggregating unique cross-system data; it compounds into a defensible foundation for future automation products.
Why it works
Broad, deep integrations create a proprietary data lake that competitors (even larger, better-funded ones) lack, and this data foundation accelerates building new AI/automation features faster than rivals.
One level up
Data breadth acquired through integration work becomes a platform asset that outlasts any single feature or product generation.
Where it stops holding
This stops working when integrations are shallow, data isn't actually used to power differentiated features, or competitors can access the same data sources equally easily.
Detail
procedure:
- Build integrations broader/deeper than competitors, including adjacent categories rivals don't touch
- Aggregate pulled data into a unified data layer ('trust graph')
- Use that data foundation to build new automation/agent products faster than competitors
benchmarks: {"integrations":"400 tools"}
rationale: In retrospect, the data foundation let Vanta build new things faster than rivals, including larger security providers.
In their words
So now you can connect Vanta to one of 400 tools. We pull a ton of data out of those tools. And that's what powers the automation, the agent, the AI, all of that. We call it our trust graph.
Model selection by quality-vs-cost tradeoff based on token intensity
Source
Invest Like the Best with Patrick O'Shaughnessy — Christina Cacioppo - Building an Enduring Software Company in the Age of AI - [Invest Like the Best, EP.496] (October 20, 2026) · at 11:09 ▸
When this applies
when deciding whether to optimize an AI-powered product for model quality or model cost
The takeaway
Let your task's token/compute intensity dictate whether you optimize for raw model quality or for cost efficiency.
Why it works
Token intensity of a task category determines whether cost or quality dominates the optimization calculus: low-token-volume domains can afford best-quality models without punishing unit economics, while high-token-volume domains (like coding) force cost sensitivity.
One level up
Match your optimization variable (quality vs. cost) to the economic structure of your workload, not to industry convention.
Where it stops holding
This stops working when task volume or token usage per task scales unexpectedly high, converting a quality-optimizable domain into a cost-sensitive one.
Detail
procedure:
- Build subject-matter-expert-authored, task-specific eval frameworks
- Eval the evals using tooling
- When swapping models, optimize primarily for quality, not cost, if your task category is low in token usage
benchmarks: {"token-heavy domains":"coding first, legal/document drafting second","token-light domain":"decision/action tasks like Vanta's"}
rationale: Because Vanta's decision-action tasks use relatively few tokens compared to coding, they can optimize for quality over cost.
In their words
When we think about model swapping, it is mostly quality. It's not cost. I also think some of it to this is because we don't use as many tokens as coding.
Only as good as your last ship
Source
Invest Like the Best with Patrick O'Shaughnessy — Christina Cacioppo - Building an Enduring Software Company in the Age of AI - [Invest Like the Best, EP.496] (October 20, 2026) · at 3:08 ▸
When this applies
when your category becomes easy for competitors to replicate because the barrier to building has collapsed
The takeaway
In a market where replication costs keep falling, sustained execution velocity becomes the moat, not historical position.
Why it works
When building cost falls (e.g., via AI code generation), first-mover advantage and existing feature sets stop being durable moats; only continuous shipping keeps you ahead.
One level up
Durable advantage must be re-earned continuously when the cost of imitation trends toward zero.
Where it stops holding
This stops working when switching costs or data/integration moats are strong enough that feature parity doesn't translate into customer switching.
Detail
rule: Treat being first as worthless; only continuous shipping of better product sustains lead.
threshold: Someone can 'vibe code your thing in a weekend' — i.e., replication time has dropped to days.
rationale: Execution and constant delivery are now the only true differentiators once building is cheap.
In their words
You're really only as good as the last thing you shipped. You can be first and you can start it. And it doesn't really matter. And there's more true now because it's like someone can vibe code your thing in a weekend.
Never pay top performers enough
Source
Invest Like the Best with Patrick O'Shaughnessy — Christina Cacioppo - Building an Enduring Software Company in the Age of AI - [Invest Like the Best, EP.496] (October 20, 2026) · at 22:44 ▸
When this applies
when deciding how aggressively to compensate your highest-value people in a market with runaway comp inflation
The takeaway
Deliberately aim to overpay your top contributors relative to market comp bands, accepting that comp dispersion should be wide and that you'll never fully 'catch up' to their value.
Why it works
High-agency, high-context people who make generally good decisions and iterate quickly create outsized value, so compensation should chase (and likely always underpay relative to) their true contribution, accepting overpayment to lower performers as a cost of the system.
One level up
In talent markets, treat top-performer compensation as an investment to maximize, not a cost to minimize to a 'fair' benchmark.
Where it stops holding
This stops working when investor tolerance for dilution/stock-based comp reverses (as happened in 2022-23 per the speaker), forcing comp growth to slow.
Detail
rule: You never pay the top people enough; always try to and you will never hit it.
threshold: Top performer total comp is roughly 10x what it was five years ago; she believes comp dispersion should be even wider.
rationale: High-agency, get-shit-done people who understand company/customers/domain and iterate from mistakes are worth 'millions of dollars' — disproportionate to typical comp bands.
In their words
My comp philosophy is: you never pay the top people enough. You should always try and you will never hit it.
Billboard moat lost due to lack of a real multi-year contract
Source
Invest Like the Best with Patrick O'Shaughnessy — Christina Cacioppo - Building an Enduring Software Company in the Age of AI - [Invest Like the Best, EP.496] (October 20, 2026) · at 49:38 ▸
When this applies
when a scarce, high-value asset underpinning your brand/identity rests on an informal or unverified agreement rather than a locked-in contract
The takeaway
When a scarce asset is core to your identity/strategy, verify and formalize the actual legal terms rather than relying on reputation or informal custom, however entrenched it appears.
Why it works
Publicly claiming a multi-year lock-up on an asset without actually securing one creates a false sense of security; when a competitor simply outbids for the same scarce asset, the informal holder loses it instantly despite being an inelastic buyer willing to pay far more.
One level up
Perceived ownership of a strategic asset is not real ownership; always verify the legal/contractual reality underlying any claimed competitive advantage.
Where it stops holding
This stops working as a warning when the asset in question is genuinely secured by formal, verified contract, or when competitors have no means or interest in outbidding you.
Detail
situation: Vanta's famous billboard location at 101/280 was assumed to be 'locked up for years' and publicly described that way, but there was no actual multi-year contract — just a purchase-order-style arrangement.
decision: When another founder offered more money for the same spot, the landlord ('policeman order') simply swapped it to the higher bidder; Vanta tried to outbid but the competitor was now cash-flow positive and inelastic as a seller of the spot.
outcome: Vanta permanently lost the iconic billboard location, which Christina described as feeling like losing real market cap/brand value.
lesson: An unverified assumption of security around a scarce strategic asset can be upended instantly by a competitor simply willing to pay more.
In their words
I think even on podcasts, I said the contract was locked up for years. It was not locked up for years.