Exaptation AI Growth Trackers
Search Signals + Growth Bets
Exaptation AI · Trackers · Release 1
Growth Search Signals
Where influential founders, CEOs, investors and accelerators say to search for growth, sorted by where each search starts.
Growth is a search problem. Every growth framework is a search, and each one looks in a different place. This tracker collects public statements about where to look for growth, uses the speakers' own words, checks each one against the original source, and sorts it into one of four frames.
Demand Pull
Also called Market Pull
The search starts from what customers or markets want.
Supply Push
The search starts from what a firm already makes or does well, pushed toward the market in its designed function.
Technology Push
The search starts from a capability that is new to the firm.
Resource Expansion
Also called Asset Expansion
Getting more value from what you already own or can access, by expanding what it does or finding functions it was never designed for.
How to read it
Each row shows who said it, the growth thesis in one sentence, a dated reason why now, the frame and the closest theory, and an ExAI note that ends with where we would start. Open any row for the source, who the advice is aimed at, the speaker's stake, and what we will check later. Sort by number, organization, source tag, frame or theory.
We are not a neutral party. We build tools for the fourth frame, so everything is in the open to disagree with and debate. We expect Resource Expansion to stay the least searched frame, and we will report where it stands at 50 signals, whatever it shows. Y Combinator's Requests for Startups are counted as their own series, so one accelerator doesn't drive the headline count.
ExAI notes are written by Shripal Shah with AI assistance and human review (Formed Mindfully with AI, or FMAI).
Read the framework behind the tracker: Growth Is a Search Problem: Four Places to Look →
Headline count · 12 signals from C-level leaders and capital allocators
1 unclassified signal, shown in the table but outside the shares. Shares exclude unclassified signals.
YC Requests for Startups series · 13 signals
Counted separately so one accelerator doesn't drive the headline.
1 Amazon · Andy Jassy C-level Build primitives, the basic building blocks; they expand your degrees of freedom, limited only by what you've built and your imagination. No dated trigger stated; Jassy presents it as a standing principle (April 2024 letter). Supply Push Resource-based view One of the few theses that points at what a firm has already built. Building blocks let others find functions their makers never designed.Where we'd start: Modularity Experiments (experiment). Rearrange existing building blocks with outside builders and watch for combinations nobody designed.
SourceCEO Andy Jassy's 2023 Letter to Shareholders (aboutamazon.com), Apr 11, 2024. Checked against the original.
Aimed atInvestors
Speaker's stakeAmazon is the example: AWS grew from this approach.
Asset angleNames assets the firm holds: its building blocks. New functions are enabled, not named.
What we'll check, and whenAmazon's later letters repeating or revising the primitives thesis (revisit Jan 31, 2027)
2 Andreessen Horowitz (a16z) · Jennifer Li Capital Allocator Unstructured enterprise data is the biggest bottleneck and the biggest untapped treasure; tools that structure it unlock enterprise AI. Li says about 80% of corporate knowledge sits in unstructured form (Dec 2025). Supply Push Resource-based view This one starts from what firms already own. The open question is whether firms find new uses for those files themselves or pay vendors to do it.Where we'd start: Resource Identification (experiment). Inventory the files first, then ask what each kind could do besides being stored.
Sourcea16z Big Ideas 2026: Part 1, Dec 9, 2025. Checked against the original.
Aimed atFounders
Speaker's stakea16z invests in data-infrastructure startups.
Asset angleNames an asset firms hold: unstructured files. Points to doing more with them, not a new function.
What we'll check, and whenEnterprise spend on unstructured-data tooling; vendors citing it in earnings (revisit Jan 31, 2027)
3 Andreessen Horowitz (a16z) · Jonathan Lai Capital Allocator World models will turn stories into co-created worlds, and those worlds will double as training grounds for AI agents and robots. A forecast for 2026, published in December 2025. Technology Push General Purpose Technology Game worlds built for play could take on a new function: training robots. That would be an emergent function for an existing asset, and studios may gain a second buyer.Where we'd start: New Application Domains (cause). Show robotics teams a studio's world assets and let them test which ones train well.
Sourcea16z Big Ideas 2026: Part 1, Dec 9, 2025. Checked against the original.
Aimed atFounders
Speaker's stakea16z invests in startups working on this.
Asset anglePoints to a new function for existing game worlds, as a forecast.
What we'll check, and whenRobotics teams licensing game-world assets for training (revisit Jan 31, 2027)
4 Andreessen Horowitz (a16z) · Sarah Wang Capital Allocator Systems of record start losing primacy in 2026 as the AI agent layer above them captures the value. A forecast for 2026, published in December 2025. Technology Push General Purpose Technology Systems of record hold years of verified business history. Before that history becomes commodity storage, its owners can ask what else it can do.Where we'd start: New Agents (cause). AI agents reading the records may find uses the original owners never saw.
Sourcea16z Big Ideas 2026: Part 1, Dec 9, 2025. Checked against the original.
Aimed atFounders
Speaker's stakea16z invests in agent-layer startups.
Asset angleNames an asset incumbents hold: systems of record, framed as losing value.
What we'll check, and whenSystem-of-record vendors' pricing power and agent-initiated workflow share in 2026 results (revisit Jan 31, 2027)
5 Microsoft · Satya Nadella C-level The most enterprise value is created during platform shifts, so lead in the AI shift. Nadella said this on Microsoft's January 2023 earnings call. Technology Push General Purpose Technology (platform shift) Platform shifts also change what existing assets can do. Data, networks and customer relationships built for the old platform can take on new functions in the new one.Where we'd start: Adjacent Possibles Innovation (experiment). With each new AI capability, re-test what existing assets can now do just beyond their current function.
SourceMicrosoft FY23 Q2 earnings conference call transcript (Microsoft Investor Relations), Jan 24, 2023. Checked against the original.
Aimed atInvestors
Speaker's stakeMicrosoft sells the AI platform.
Asset angleA when, not a what. No asset named.
What we'll check, and whenMicrosoft commentary on value capture during the AI platform shift in 2026 calls (revisit Jan 31, 2027)
6 Nvidia · Jensen Huang C-level Within a decade, about 75,000 Nvidia employees will work alongside 7.5 million AI agents doing the grunt work. Huang said this at Nvidia's GTC conference in March 2026. Technology Push General Purpose Technology Millions of agents will need work to do. Pointing some at a firm's existing assets, looking for functions nobody designed, could be one of their more valuable uses.Where we'd start: Thought Experiments (experiment). Give some agents a standing prompt: what else could this asset do?
SourceFortune, reporting Huang's GTC 2026 press Q&A in San Jose, Mar 19, 2026. Checked against a report quoting the speaker directly.
Aimed atIncumbents
Speaker's stakeNvidia sells the computing that AI agents run on.
Asset angleNo asset named; agents are aimed at designed tasks.
What we'll check, and whenNvidia disclosures on agents per employee (revisit Jan 31, 2027)
7 OpenAI · Sam Altman C-level Coding agents now do in minutes what took a startup three months, so founders should take on far more ambitious projects, including hard technology. YC's Garry Tan, in the same talk: hard tech rose from 5 to 10% of batches to 15 to 25%. Technology Push General Purpose Technology When building gets this cheap, the scarce input shifts to what a company already holds: data, relationships and physical assets. Ambitious projects that start from those are harder to copy.Where we'd start: First Principles Engineering (experiment). List primitives for data, relationships and physical assets you hold, then reason up to find novelty.
SourceSam Altman: 'Never a Better Time to Do a Startup' (YC Startup School 2026, in conversation with Garry Tan), Jul 28, 2026. Checked against the original.
Aimed atFounders
Speaker's stakeOpenAI sells the AI models and coding agents founders would build with.
Asset angleStarts from a new technology. No asset named.
What we'll check, and whenHard-tech share of YC batches; outcomes of ventures built largely by agents (revisit Feb 11, 2027)
8 OpenAI · Sam Altman C-level Look for ideas where you can build real conviction that conventional wisdom is wrong, and accept that it will take a long time. Altman's example: ten years ago many experts called trying to build AGI irresponsible (Jul 2026 talk). Unclassified Folk heuristic (contrarian bet) Assets everyone knows by one function are a natural home for this. The belief that a thing can only do what it was built for is conventional wisdom too.Where we'd start: Forward Search Experiments (experiment). Push an asset into conditions experts say it can't handle and record what it does.
SourceSam Altman: 'Never a Better Time to Do a Startup' (YC Startup School 2026, in conversation with Garry Tan), Jul 28, 2026. Checked against the original.
Aimed atFounders
Speaker's stakeOpenAI sells the AI models and coding agents founders would build with.
Asset angleA way of choosing where to look. No asset named.
What we'll check, and whenWhich of today's dismissed ideas gain funding or users within two years (revisit Feb 11, 2027)
9 Stripe · Patrick Collison C-level Build the good version of the clunky 1990s software that still runs whole industries. YC asked founders in July 2026 to replace 20-year-old field-service software. Demand Pull General demand pull The software being replaced sits on years of customer and work records. The open question is who puts that history to new use first, the newcomer or the incumbent.Where we'd start: Edge Case Portfolio (experiment). Old industry software is full of user workarounds. Some point to functions it was never built for.
SourceThe Best Time in History to Start a Company (Collison with Amjad Masad), YouTube, May 22, 2026. Checked against the original.
Aimed atFounders
Speaker's stakeStripe earns when more companies start and take payments.
Asset angleStarts from incumbents' weak software, not from an asset the reader holds. No new function named.
What we'll check, and whenMore funded startups replacing legacy vertical software; churn at incumbents in named verticals (revisit Jan 31, 2027)
10 Stripe · Patrick Collison C-level Look for whole sectors software has barely touched; domain experts are already building niche tools, like software for ice rinks. Collison calls this the best time in history to start a company (May 2026). Demand Pull General demand pull Untouched sectors are rarely empty. Equipment makers, distributors and trade groups already hold the customer lists and know-how a newcomer has to build.Where we'd start: Niche Exploration (cause). Map who already serves the sector and what their tools quietly do beyond their designed function.
SourceThe Best Time in History to Start a Company (Collison with Amjad Masad), YouTube, May 22, 2026. Checked against the original.
Aimed atFounders
Speaker's stakeStripe earns when more companies start and take payments.
Asset angleStarts from an underserved market. Assets that nearby incumbents hold are not mentioned.
What we'll check, and whenNiche-vertical startups built by domain experts reaching meaningful revenue (revisit Jan 31, 2027)
11 Stripe · Patrick Collison C-level Watch what is popular with young people but still low status; incumbents tend to look down on it. No dated trigger stated; the advice is evergreen. Demand Pull Disruptive Innovation Low-status use is often where people repurpose tools before anyone designed them for it. Watching how young users bend existing products is a cheap early signal.Where we'd start: Bricolage (cause). Watch how young users repurpose existing products for functions their makers never planned.
SourceThe Best Time in History to Start a Company (Collison with Amjad Masad), YouTube, May 22, 2026. Checked against the original.
Aimed atFounders
Speaker's stakeStripe earns when more companies start and take payments.
Asset angleStarts from behavior. Whether it points to new functions for existing tools is not yet clear.
What we'll check, and whenA low-status youth behavior becoming a funded category within 12 months (revisit Jan 31, 2027)
12 Stripe · Patrick Collison C-level Start trying things right away instead of waiting for a fully formed idea. Stripe's 2025 letter: 20% of Atlas startups charged a first customer within 30 days, up from 8% in 2020. Demand Pull Lean Startup Fast trials work best when they start from what you already have. A quick test of whether an existing asset can perform a new function costs little and teaches a lot.Where we'd start: Constraint-Driven Reframing Test (experiment). Give a team one week and only existing assets to solve a new problem.
SourceThe Best Time in History to Start a Company (Collison with Amjad Masad), YouTube, May 22, 2026. Checked against the original.
Aimed atFounders
Speaker's stakeStripe earns when more companies start and take payments.
Asset angleA way of searching, not a place to look. No asset or new function named.
What we'll check, and whenStripe updates on time to first revenue for new companies (revisit Jan 31, 2027)
13 US Army · Daniel P. Driscoll YC RFS Fall 2026 The Army is actively funding low-cost interceptors, sensors, drones and logistics tools that plug into its open system architecture. Driscoll says the Army has rewritten its acquisition playbook (Jul 2026). Demand Pull General demand pull Products built for other markets, from industrial sensors to commercial drones, may already meet part of this need. An open architecture lowers the cost of trying.Where we'd start: Structural Borrowing Experiments (experiment). Test which components of commercial products already meet a defense need inside the open architecture.
SourceYC Requests for Startups, Fall 2026, Jul 2026. Checked against the original.
Aimed atFounders
Speaker's stakeThe Army is the buyer of what it requests.
Asset angleImplies an asset others must plug into: the Army's open architecture.
What we'll check, and whenArmy awards to startups through the new acquisition path; reported cost-per-kill reductions (revisit Jan 31, 2027)
14 Y Combinator · Aaron Epstein YC RFS Fall 2026 AI is still single-player; as agents run tasks lasting hours or weeks, teams need shared sessions they can watch and redirect. YC says agents now run tasks that take hours or days (Jul 2026). Demand Pull General demand pull Google Docs and Figma won by going multiplayer. Firms that already host shared workspaces start with the users and habits this needs.Where we'd start: Stakeholder-Led Innovation (experiment). Put a whole team, not one user, in front of an agent and watch which shared uses emerge.
SourceYC Requests for Startups, Fall 2026, Jul 2026. Checked against the original.
Aimed atFounders
Speaker's stakeYC funds startups that answer its requests, at $500,000 each.
Asset angleStarts from how teams work. No asset named.
What we'll check, and whenMultiplayer agent products launched by major vendors or startups (revisit Jan 31, 2027)
15 Y Combinator · Andrew Miklas YC RFS Fall 2026 Build an AI tutor that teaches young children to read, write and count at private-tutor quality and consumer scale. YC says this kind of tutor is starting to feel possible for the first time (Jul 2026). Technology Push General Purpose Technology Publishers and schools already hold decades of graded children's work and lesson material. That record could shape what a good tutor looks like.Where we'd start: User-Led Function Discovery Sprint (experiment). Let children and teachers bend early versions and keep the uses that stick.
SourceYC Requests for Startups, Fall 2026, Jul 2026. Checked against the original.
Aimed atFounders
Speaker's stakeYC funds startups that answer its requests, at $500,000 each.
Asset angleStarts from a new technology. No asset named.
What we'll check, and whenConsumer AI tutors for young children reaching scale with measured learning gains (revisit Jan 31, 2027)
16 Y Combinator · Austin Tindle and Diana Hu YC RFS Fall 2026 Dense physical-world data is now cheap enough to collect, so build models that describe, then control, physical systems. Sorcerer's weather-balloon data now reaches NOAA through the National Mesonet Program. Technology Push Specific technology Mobile networks already log signal strength that can measure rainfall, as shown in Science in 2006. Some of the densest data may already sit in operators' logs.Where we'd start: Data Anomaly Analysis (experiment). Signal drops that engineers treat as noise can carry a measurement of the physical world.
SourceYC Requests for Startups, Fall 2026, Jul 2026. Checked against the original.
Aimed atFounders
Speaker's stakeTindle founded Sorcerer, the request's own example; YC funds startups that answer its requests.
Asset angleImplies assets incumbents hold: sensor logs from existing networks.
What we'll check, and whenMore sensor-data startups selling to government or industry buyers (revisit Jan 31, 2027)
17 Y Combinator · Charlie Warren YC RFS Fall 2026 Field industries need software that routes work across AI agents, robots and people, replacing 20-year-old dispatch and billing tools. Warren says these industries spend 10 to 100 times more on labor than on software. Technology Push General Purpose Technology The startups' stated moat is a record of work as it happens. Field-service firms already keep years of work records that could train and test those agents.Where we'd start: Data Science Experiments (experiment). Test whether years of work records can predict service times well enough to train routing agents.
SourceYC Requests for Startups, Fall 2026, Jul 2026. Checked against the original.
Aimed atFounders
Speaker's stakeYC funds startups that answer its requests, at $500,000 each.
Asset angleImplies an asset incumbents hold: work records.
What we'll check, and whenFunded YC companies answering this request; incumbents adding agent routing (revisit Jan 31, 2027)
18 Y Combinator · Daivik Goel YC RFS Fall 2026 Financial compliance runs on spreadsheets, point tools and headcount; build AI-native infrastructure that consolidates it with real-time visibility. Goel says compliance costs grow faster than revenue as firms expand into new markets (Jul 2026). Demand Pull General demand pull Compliance teams already hold years of filings, findings and regulator letters. That record is the raw material for any tool that learns to stay compliant.Where we'd start: Reusability Experiments (experiment). Reuse past filings and regulator letters as test cases for every new market entry.
SourceYC Requests for Startups, Fall 2026, Jul 2026. Checked against the original.
Aimed atFounders
Speaker's stakeYC funds startups that answer its requests, at $500,000 each.
Asset angleStarts from a customer problem. Compliance records are implied.
What we'll check, and whenAI compliance tools replacing point solutions; reported cost and headcount effects (revisit Jan 31, 2027)
19 Y Combinator · Francois Chaubard YC RFS Fall 2026 Data centers are short of power, land and permits; move compute offshore onto modular vessels using the ocean's sunlight and cooling. YC says new data centers can take years to approve and can still be blocked locally (Jul 2026). Resource Expansion Exaptation Growth Strategy This puts an existing resource, the ocean, to work as a cooling system. Shipyards and offshore operators already hold the skills and vessels it needs.Where we'd start: Simulations with Resource Properties (experiment). Model how vessels, seawater and sunlight behave together before building the first floating data center.
SourceYC Requests for Startups, Fall 2026, Jul 2026. Checked against the original.
Aimed atFounders
Speaker's stakeYC funds startups that answer its requests, at $500,000 each.
Asset angleStarts from a natural resource's properties. Whether this is a new function is unclear.
What we'll check, and whenOffshore compute pilots; earlier underwater data center trials revisited (revisit Jan 31, 2027)
20 Y Combinator · Harsha Gaddipati YC RFS Fall 2026 API providers should not just announce changes but apply them, with agents that open fixes in customers' code. Gaddipati says over 30% of service downtime he saw at AWS came from unnoticed external API changes. Technology Push General Purpose Technology API vendors already hold their own change logs and usage data. Turning a changelog from a notice into a fix is a new function for an old record.Where we'd start: Programmable Products and Services (experiment). Turn changelogs into instructions agents can run, then see what else those instructions can do.
SourceYC Requests for Startups, Fall 2026, Jul 2026. Checked against the original.
Aimed atFounders
Speaker's stakeYC funds startups that answer its requests, at $500,000 each.
Asset angleImplies assets vendors hold: change logs and usage data.
What we'll check, and whenAPI vendors shipping update agents; adoption of a neutral cross-vendor service (revisit Jan 31, 2027)
21 Y Combinator · Max Kolysh YC RFS Fall 2026 The aging population needs care that caregivers can't supply; AI now enables voice companions, safety monitoring, home robots and care coordination. By 2030 one in five Americans will be over 65; 53 million family members already give unpaid care. Demand Pull General demand pull Homes are already full of devices, from speakers to thermostats, that could watch over an older person. Some of the answer may be new functions for things already installed.Where we'd start: Shift of Function (experiment). Test whether sensors already in the home, built for comfort or music, can notice a fall.
SourceYC Requests for Startups, Fall 2026, Jul 2026. Checked against the original.
Aimed atFounders
Speaker's stakeYC funds startups that answer its requests, at $500,000 each.
Asset angleStarts from customer need. No asset named.
What we'll check, and whenAdoption of AI caregiving products; caregiver shortage data (revisit Jan 31, 2027)
22 Y Combinator · Max Kolysh YC RFS Fall 2026 Cheap deepfakes have broken old trust signals; rebuild the internet's trust layer so every call and payment confirms a real human. A finance worker wired out $25 million after a video call where everyone else was a deepfake. Demand Pull General demand pull Phone numbers, emails and ID numbers took on verification functions they were never designed for. This request says those borrowed functions are now failing.Where we'd start: Need-Solution Pairs (cause). The next verification signal may already exist somewhere, waiting to be paired with this new need.
SourceYC Requests for Startups, Fall 2026, Jul 2026. Checked against the original.
Aimed atFounders
Speaker's stakeYC funds startups that answer its requests, at $500,000 each.
Asset angleStarts from a customer problem. Existing identifiers are the assets under strain.
What we'll check, and whenVerified-human products adopted by banks or major platforms (revisit Jan 31, 2027)
23 Y Combinator · Nemil Dalal YC RFS Fall 2026 A crypto bear market is the best time to build; YC expects every startup to use crypto rails eventually. YC cites clearer regulation and stablecoins spreading through major institutions (Jul 2026). Technology Push Specific technology The firms running today's payment systems hold the customer trust these rails still need, which makes their next move worth watching.Where we'd start: Catalytic Reactors (cause). Ask which existing assets could be financed or traded in new ways once they run on these rails.
SourceYC Requests for Startups, Fall 2026, Jul 2026. Checked against the original.
Aimed atFounders
Speaker's stakeYC funds startups that answer its requests, at $500,000 each.
Asset angleStarts from a technology. No asset named.
What we'll check, and whenStablecoin adoption by major institutions; count of YC crypto investments (revisit Jan 31, 2027)
24 Y Combinator · Pete Koomen YC RFS Fall 2026 Agents make one-team tools cheap to build; what's missing is a cloud that makes them as easy to share as a Google Doc. Koomen says agents have made writing personal software nearly free (Jul 2026). Technology Push General Purpose Technology The hard part is trust: sign-in, permissions and sharing. Workspace vendors and corporate IT already run all three for every document employees touch.Where we'd start: Interoperability Experiments (experiment). Open existing sign-in and sharing systems to agent-built tools and see what new uses appear.
SourceYC Requests for Startups, Fall 2026, Jul 2026. Checked against the original.
Aimed atFounders
Speaker's stakeYC funds startups that answer its requests, at $500,000 each.
Asset angleImplies an asset incumbents hold: permissions and sharing systems.
What we'll check, and whenLaunch of small-software hosting by startups or workspace vendors (revisit Jan 31, 2027)
25 Y Combinator · Raphael Schaad YC RFS Fall 2026 Every platform shift mints consumer giants; AI costs per user are falling about 10x a year, so build consumer products now. YC says AI's cost per user is falling about 10x a year (Jul 2026). Technology Push General Purpose Technology (platform shift) Past consumer giants often grew from something users already had, like a phone's camera or contacts. Cheap AI could give existing devices and data new functions too.Where we'd start: Spotting Emergent Patterns (cause). Watch what people already do with AI on their phones, then build for the habits that repeat.
SourceYC Requests for Startups, Fall 2026, Jul 2026. Checked against the original.
Aimed atFounders
Speaker's stakeYC funds startups that answer its requests, at $500,000 each.
Asset angleStarts from a new technology. No asset named.
What we'll check, and whenNew consumer AI apps reaching mass adoption; per-user cost trend (revisit Jan 31, 2027)
Theory labels are our reading unless the speaker named the theory. Frames are provisional and may change as the tracker grows. Every signal is checked against the original source, or a report quoting the speaker directly, before it is published.
We are building the search engine for every asset's undiscovered functions.