ExpertsFMDara Khosrowshahi - Uber's Bet on AVs, AI, and Building a Super-App - [Invest Like the Best, EP.476]
Invest Like the Best with Patrick O'Shaughnessy · June 3, 2026
5 Selected Lessons:
framework · 3 reasoning_pattern · 1 heuristic · 1
Bottoms-up AI adoption with first-principles rebuild
Source
Invest Like the Best with Patrick O'Shaughnessy — Dara Khosrowshahi - Uber's Bet on AVs, AI, and Building a Super-App - [Invest Like the Best, EP.476] (June 3, 2026) · at 16:29 ▸
When this applies
when deploying a transformative tool across an organization with many different functions and unpredictable early adopters
The takeaway
For transformative technology, decentralize experimentation to surface unpredictable wins, then intervene specifically to push teams from incremental optimization toward full process redesign.
Why it works
A top-down mandate can't predict where the technology will have outsized impact; letting adoption emerge bottoms-up surfaces unpredictable high-leverage use cases (e.g., 10x code commits from unexpected teams), while leadership's role shifts to pushing first-principles rebuilds rather than incremental optimization.
One level up
New general-purpose capabilities are best exploited through distributed experimentation rather than centralized planning.
Where it stops holding
This stops working when the technology's safe/effective use requires centralized expertise or compliance controls that individual teams can't be trusted to self-govern.
Detail
procedure:
- Avoid single top-down mandate; keep a bottoms-up invention culture
- Let AI adoption emerge organically across all functions (engineering, legal, marketing)
- Identify unpredictable high performers/rebels and promote them
- Push teams toward full first-principles process redesign, not just 20-30% optimization of an existing process
benchmarks: {"code_commits_multiplier":"devs in India driving 10x the code commits"}
rationale: It could be a decent first step to optimize 20-30% of a process, but rebuilding it from the ground up is much harder and more valuable.
In their words
It could be a decent first step to take a process and have AI optimize 20% of it, 30% of it in terms of speed or effectiveness or cost. But to rebuild it from the ground up, that's much harder.
Fixed-cost vs variable-cost membership design
Source
Invest Like the Best with Patrick O'Shaughnessy — Dara Khosrowshahi - Uber's Bet on AVs, AI, and Building a Super-App - [Invest Like the Best, EP.476] (June 3, 2026) · at 42:25 ▸
When this applies
when designing any loyalty/subscription program with per-use benefits
The takeaway
Classify a membership program by whether its benefits draw on fixed or variable cost inventory, because that determines whether profitability shows up immediately (fixed-cost case) or only over the customer's lifetime (variable-cost case), and plan your tolerance for near-term losses accordingly.
Why it works
Ideal membership programs monetize fixed-cost or near-zero-marginal-cost inventory (unused hotel rooms, content library), so the only real cost is acquisition; variable-cost membership programs (where benefits scale with usage, as with Uber One's discounts) are harder because heavy users are structurally less profitable per transaction, requiring a lifetime-value bet instead of per-transaction profitability.
One level up
Durable business model innovations sometimes require accepting a multi-year 'valley of despair' that public markets misread as failure, if the underlying unit economics genuinely improve with tenure.
Where it stops holding
This stops working when customer tenure/retention is too short for the later high-margin years to ever arrive, in which case the early losses are never recovered.
Detail
procedure:
- Identify if membership benefits draw from fixed-cost assets (ideal) or variable-cost services (harder)
- If variable-cost, model lifetime value: accept lower/negative profitability in year one
- Track profitability improving in years two, three, four as members' habits form
- Stay the course through public skepticism if unit economics trend positively
benchmarks: {"membership_scale":"50 million Uber One members, growing 50% YoY","profitability_timeline":"member is less profitable in year one, solidly profitable now at platform level"}
rationale: Amazon walked through a 'valley of despair' with Prime because they understood underlying unit economics even as public markets misread rising losses.
In their words
Over the lifetime, you will be more profitable. And we're seeing that.
Seeking organizational troublemakers as mutation sources
Source
Invest Like the Best with Patrick O'Shaughnessy — Dara Khosrowshahi - Uber's Bet on AVs, AI, and Building a Super-App - [Invest Like the Best, EP.476] (June 3, 2026) · at 56:53 ▸ · at 59:43 ▸
When this applies
when a growing organization risks losing its capacity to adapt because of cultural homogenization and rigid processes
The takeaway
Deliberately seek out and protect internal dissenters/unconventional thinkers, and build unstructured, random interactions into your schedule, because scaled organizations naturally suppress the variation needed to adapt.
Why it works
As organizations scale, incentives push toward conformity and smooth processes, which suppress the variation (dissenting ideas, unconventional people) that adaptation requires; deliberately protecting and promoting 'troublemakers' reintroduces that variation, like genetic mutation in an evolving organism.
One level up
Organizations, like organisms, require a constant source of 'mutation' (dissent, unconventional ideas) to avoid stagnation and eventual death.
Where it stops holding
This stops working when 'troublemaking' is indulged without discernment, creating disruptive noise rather than useful signal, or when the organization lacks mechanisms to act on strong signals once found.
Detail
procedure:
- Build deliberately random, unstructured interactions into your schedule (not just direct reports)
- Actively look for and bring in 'troublemakers' who challenge the single dominant culture/process
- Take in signal even if you don't act immediately
- Act once the signal becomes strong enough
benchmarks: {}
rationale: Companies that don't mutate, that sit with a single process and information flow, are the companies that die.
In their words
Organisms evolve by mutating. And companies that don't mutate, that just sit and with a single process, with a single information flow, those are the companies that die.
Coexistence model for platform vs. direct-channel partners
Source
Invest Like the Best with Patrick O'Shaughnessy — Dara Khosrowshahi - Uber's Bet on AVs, AI, and Building a Super-App - [Invest Like the Best, EP.476] (June 3, 2026) · at 26:17 ▸
When this applies
when a platform's suppliers are simultaneously potential competitors building their own direct-to-consumer channel
The takeaway
Relationships between aggregators and asset-owning suppliers are rarely binary compete-or-cooperate; both sides can gain simultaneously because the aggregator drives marginal utilization the supplier's own channel can't fully capture.
Why it works
Suppliers value incremental utilization of their fixed asset base (hotel rooms, AV fleets) even while building their own loyalty channel, because marginal utilization (e.g., 70% vs 90% occupancy) is extremely valuable, so both relationships can coexist without being purely competitive or purely cooperative.
One level up
In markets with fixed, perishable capacity, aggregation and direct-channel building are complementary, not mutually exclusive, strategies.
Where it stops holding
This stops working when the supplier's own direct channel becomes large enough that the marginal utilization gain from the aggregator no longer outweighs the competitive threat of losing customer relationships.
Detail
approach: Analogize new partner dynamics (AV providers) to a known precedent (OTAs and hotels/airlines) to predict how competition and cooperation will coexist
signals:
- partner has its own fixed, perishable capacity
- partner wants both its own channel and incremental aggregator-driven demand
rationale: The returns for a hotel filled to 90% vs 70% are night and day — they'll take the incremental demand even while competing for the direct relationship.
In their words
The returns for a hotel that has 70% of its rooms filled versus 90%, it's nine and day. You will take the 90% every single day.
Go to the unfiltered source for ground truth
Source
Invest Like the Best with Patrick O'Shaughnessy — Dara Khosrowshahi - Uber's Bet on AVs, AI, and Building a Super-App - [Invest Like the Best, EP.476] (June 3, 2026) · at 53:10 ▸ · at 56:53 ▸
When this applies
when a leader's information is being filtered by layers of staff before it reaches them
The takeaway
Deliberately bypass processed information and go to the person/data closest to the raw work, because the edge cases filtered out along the way are where decision-making advantage lives.
Why it works
Information gets smoothed and averaged as it passes through organizational layers, stripping out the outlier ('edge') details that actually drive differentiated decisions, since everyone reacts similarly to the average but the edge case is where advantage lies.
One level up
The higher you rise in an organization, the more your information is pre-processed for you; seeking unfiltered truth requires deliberate, sometimes uncomfortable effort.
Where it stops holding
This stops working when going around the hierarchy undermines trust or accountability with the managers who were bypassed, or when the 'source' lacks broader context the filtered summary would have provided.
Detail
rule: Insist on talking to the person who actually did the work, not the manager summarizing it; build deliberate randomness into your schedule to counteract processed information
threshold: Applies especially for high-stakes decisions (e.g., committing billions in debt)
rationale: It's the filtering that gets the edge out of the story; it's often the edge that gives you an edge, not the average.
In their words
It's the filtering that gets the edge out of the story or out of the situation. And it's often the edge that gives you an edge, not the average.