ExpertsFM20VC: Meta's Muse Hits No. 1. ChatGPT Finally Has a Rival | Menlo Sounds the AI Bubble Alarm | Factory Triples Its Valuation to $5 Billion | Keith Rabois vs Airwallex: Who is Right? | Crusoe's $3.9 Bi
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch · September 24, 2026
5 Selected Lessons:
reasoning_pattern · 2 heuristic · 2 case · 1
Mapping platform incentives to predict who blocks vs. embraces AI agents
Source
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch — 20VC: Meta's Muse Hits No. 1. ChatGPT Finally Has a Rival | Menlo Sounds the AI Bubble Alarm | Factory Triples Its Valuation to $5 Billion | Keith Rabois vs Airwallex: Who is Right? | Crusoe's $3.9 Bi (September 24, 2026) · at 17:01 ▸ · at 20:19 ▸
When this applies
when deciding whether to open your platform's API to an aggregator/agent that could disintermediate your existing revenue model
The takeaway
To predict a platform's reaction to disintermediating technology, check whether its profit pool depends on the surface being bypassed, and how much bargaining leverage it has over the aggregator.
Why it works
A platform's decision to block or embrace an intermediary is explained by whether that platform's profit depends on controlling the surface the agent would bypass (e.g. ads, upsell) and by its relative leverage/power versus the agent.
One level up
Executives act rationally according to what maximizes their own economics, not out of abstract tech ideology.
Where it stops holding
This stops working when regulatory or reputational pressures override pure economic calculation, or when the agent-driven demand becomes so large that leverage dynamics flip quickly.
Detail
approach: Identify what revenue stream the platform depends on (ads, upsell, basket size) and whether it has enough market power to force agents to come through it anyway.
signals:
- ad revenue exceeding core profit (Amazon)
- reduced basket size from bypassing browsing (Amazon/Walmart)
- no significant ad business to protect (Shopify)
- payment rail monetization as an alternative lever (Shopify)
rationale: Each platform's board-level incentive structure, not ideology, determines its agent policy; capitalism makes these decisions logical and predictable.
In their words
capitalism works. These executives are very logical. They're like, it's great that you have this new source of demand. What matters to me? In Amazon's case, I think I have a lot of leverage.
Disaggregating risk across repriced rounds for the same company
Source
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch — 20VC: Meta's Muse Hits No. 1. ChatGPT Finally Has a Rival | Menlo Sounds the AI Bubble Alarm | Factory Triples Its Valuation to $5 Billion | Keith Rabois vs Airwallex: Who is Right? | Crusoe's $3.9 Bi (September 24, 2026) · at 41:59 ▸ · at 45:24 ▸ · at 49:55 ▸
When this applies
when evaluating whether to invest in later rounds of a company whose valuation has jumped dramatically in a short time
The takeaway
Never treat a founder's or fund's early win as proof that betting equally aggressively on their much later, much pricier rounds carries the same risk — price itself changes the bet.
Why it works
Each round at a materially different valuation carries a materially different risk-return profile even for the identical underlying business, because the price paid determines the margin of safety against downside scenarios.
One level up
Conviction in a team or thesis does not transfer uniformly across price points; risk is a function of price paid relative to intrinsic uncertainty, not of identity.
Where it stops holding
This stops working when the company's fundamentals have also derisked proportionally to the price increase (e.g. clear path to an even larger acquisition), making the higher price still justified.
Detail
approach: List each round's valuation and ask whether the margin of safety (per Buffett) at that specific price still justifies the same conviction.
signals:
- valuation jumping 20x+ within months
- same underlying business and team across rounds
- temptation to extrapolate early-round success to justify later, pricier rounds
rationale: Margin of safety shrinks mechanically as price rises, independent of the quality of the team or idea.
In their words
you cannot say the risk in all four of them is the same. Because one of them is, let's say 20 yeah 20 times more expensive.
The internet (and AI) 'abhors' middlemen whose only value is reduced search cost
Source
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch — 20VC: Meta's Muse Hits No. 1. ChatGPT Finally Has a Rival | Menlo Sounds the AI Bubble Alarm | Factory Triples Its Valuation to $5 Billion | Keith Rabois vs Airwallex: Who is Right? | Crusoe's $3.9 Bi (September 24, 2026) · at 17:01 ▸ · at 20:19 ▸
When this applies
when a new technology can close the information gap that a middleman business was monetizing
The takeaway
Middlemen whose entire value is information arbitrage get compressed whenever a cheaper, tireless way to match supply and demand appears.
Why it works
If a business's value proposition is purely reducing search/matching costs between supply and demand, any technology that reduces those costs further (like a tireless agent hitting every API) erodes that business's reason to exist.
One level up
Efficiency-seeking technologies systematically squeeze pure intermediation out of markets over time.
Where it stops holding
This stops working when the middleman also provides irreplaceable physical infrastructure (fulfillment, delivery, inventory) that the new technology can't substitute, which is why full business destruction ('killed') is less likely than partial margin erosion ('maimed').
Detail
rule: If even 5-10% of an aggregator's ad/intermediation revenue is displaced by agents, that's a meaningful (though not fatal) hit.
threshold: 5-10% revenue impact is enough to be 'meaningful'; full replacement ('killed') requires losing the physical/infrastructure advantage too.
rationale: Agents will gleefully find broken/open APIs to route around exclusive gatekeepers, since they don't tire or care about friction the way humans do.
In their words
the internet abhors inefficiency. In other words, middlemen get pounded down, right?
LP seed/Series A allocation: chase the best deals, not ownership percentage
Source
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch — 20VC: Meta's Muse Hits No. 1. ChatGPT Finally Has a Rival | Menlo Sounds the AI Bubble Alarm | Factory Triples Its Valuation to $5 Billion | Keith Rabois vs Airwallex: Who is Right? | Crusoe's $3.9 Bi (September 24, 2026) · at 49:55 ▸
When this applies
when an allocator must decide whether to prioritize ownership stake or access to the best opportunities
The takeaway
At the earliest, smallest-check stages, prioritize access to the most exciting opportunities over maximizing ownership percentage, because access compounds into ownership advantages later.
Why it works
A seed investor who consistently takes smaller ownership but gets into the very best deals builds a reputation that earns them more capital and access over time, compounding into larger ownership later; chasing ownership percentage directly can mean settling for worse deals.
One level up
Optimize for the scarce resource (access/quality) rather than the visible metric (ownership %) when they trade off.
Where it stops holding
This stops working at later growth stages where capital deployed is large relative to opportunity size, at which point ownership percentage and valuation discipline (margin of safety) become the binding constraint, per Rory's own distinction between seed/Series A and later rounds.
Detail
rule: For seed/Series A LP allocation, prioritize managers in the best new high-growth opportunities over managers who insist on high ownership.
threshold: Target ~10% ownership at Series A, 15-20% at seed, but don't screen managers in/out purely on that basis.
rationale: A seed investor who consistently shows up with smaller ownership but in the best deals gets more capital and more opportunities over time.
In their words
a seed investor who consistently shows up, as Jason will attest, with smaller ownership, but in the best deals will get more ownership over time because they'll get more capital and get more opportunities.
Jason's hands-on struggle with model-routing harnesses
Source
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch — 20VC: Meta's Muse Hits No. 1. ChatGPT Finally Has a Rival | Menlo Sounds the AI Bubble Alarm | Factory Triples Its Valuation to $5 Billion | Keith Rabois vs Airwallex: Who is Right? | Crusoe's $3.9 Bi (September 24, 2026) · at 23:50 ▸ · at 30:02 ▸
When this applies
when adopting any automated decision-routing system for mission-critical tasks where the tool is right most but not all of the time
The takeaway
Adopting an automated router shifts the burden from making decisions to continuously validating the router's decisions — a cost that scales with the number of use cases and the criticality of errors.
Why it works
A routing tool with even a modest error rate (e.g. ~20%) on consequential decisions creates downstream costs (debugging, QA, lost trust) that can outweigh its speed/cost savings unless rigorously tested per use case.
One level up
Automation that trades judgment for speed still requires ongoing human oversight proportional to the stakes of being wrong.
Where it stops holding
This stops working when the task's error tolerance is low (mission-critical coding) rather than high (casual matching recommendations), where even a well-performing router becomes unusable without heavy custom QA.
Detail
situation: Jason tried Jev for deciding who should meet whom in the SaaS community, and tried an auto-router for coding model selection.
decision: Invested time to rebuild prompts; found Jev fast and cheap for the narrow matching use case but wrong ~20% of the time on more complex judgment calls, and reverted the coding auto-router back to a single trusted model after performance degradation.
outcome: The tool worked brilliantly for low-stakes judgment but was unreliable enough on higher-stakes tasks that Jason 'tapped out' of trying to manage model selection himself.
lesson: Automated model-routing is genuinely valuable but dramatically increases the operational burden of evals/QA rather than simplifying development, especially as error rates (even ~20%) compound on mission-critical work.
In their words
Jev is like wrong like 20% of the time. If you pick the wrong model for me coding a mission-critical feature 20% of the time, I'm going to be trying to build this thing bug-ridden all day long.