Gatekeeper collapse
APIs blew open closed systems; AI collapses the integration lag. Here’s where value goes next.
A solo founder just deployed an AI voice‑bot that fields more calls than a 300‑seat BPO. I grew up in Bangalore during the peak of the outsourcing boom. Back then, you needed carrier contracts, racks of hardware, and a floor of agents. Last weekend, she needed an API key. That jump—from months of negotiation to minutes of code—is what a gate falling looks like.
This keeps happening. I’ve seen it from the inside. At Stripe and TrueLayer, we turned things that sat behind the counter into API calls. Place a phone call. Charge a card. Now: use a language model. Wrap the gated thing into a self-serve interface and anyone can build on top. Telcos, banks, PhDs who ran the counter no longer control who builds on top.
That’s gatekeeper collapse: the flip from negotiated access to programmatic access.
Once that flip happens, usage explodes, and the scarce thing moves.
What makes something gated?
I don’t have a perfect definition here. A capability is gated when only a narrow, accredited, or heavily capitalized group can use it.
The three common types of gates:
Permission: do you need approval from someone powerful (regulator, telco, bank, broker‑dealer) before you can ship? In 2007 you couldn’t launch a text‑messaging app without telcos approving short codes. In 2012 you needed a broker‑dealer relationship to trade stocks programmatically.
Coordination: does it work only after lots of other people show up? Credit cards are the canonical example. A single merchant with a terminal is useless; so is a single cardholder. You need both sides at scale before the network clears.
Capital (time counts): must you spend millions—or wait a year that behaves like money—before the thing even runs? Think semiconductor fabs, nationwide 5G, or GPU cluster.
Rule of thumb: If hundreds of orgs can do it, not hundreds of thousands, it’s gated.
What happens when a gate falls?
Before: you begged, filled forms, raised capital, waited.
Pre‑Twilio you begged telcos for numbers. Post‑Twilio a curl call rang a phone. Twilio booked $4.46B in revenues and served more than 325K active accounts in 2024.
Pre‑Stripe you prayed for a merchant ID. Post‑Stripe two lines of JS took cards. Stripe processed $1.4T of volume in 2024 (~1.3% of global GDP).
Pre‑OpenAI you hired PhDs and bought GPUs. Post‑OpenAI, you prompt. OpenAI hit a ~$10B annualized revenue run rate (June 2025), up from ~$5.5B (Dec 2024) as API and ChatGPT usage exploded.
When a gate drops you see two waves: (1) a flood of builders; (2) demand for what’s still hard (uptime, fraud, compliance, cost, domain fit).
Platforms that smash the gate often sell those higher‑order things—but not all of them—so new companies emerge just above.
Two ways to win after collapse
The ecosystem that forms around a fallen gate has two types of players.
Gate‑Smasher Platforms. Expose the thing. Clean API, messy regulated/capital‑heavy backend. Monetize usage; later upsell tools.
Twilio (telephony)
Stripe (payments)
OpenAI API (models)
GroqCloud (tokens / fast inference)
Post‑Collapse Specialists. Sell whatever is scarce once access is cheap: trust, observability, orchestration, domain data, margin control.
Datadog (observability once compute got cheap)
Terraform Cloud (infra orchestration)
Stripe Radar & SentiLink (fraud/risk after payments got easy)
Most great API companies do both over time: first open the gate; later solve the higher‑level pain.
APIs x AI: The double whammy (Why gates fall fast)
Most AI capabilities arrive as APIs. APIs already let new capabilities spread fast. But AI makes them spread faster: the model itself can call other APIs, write the glue, and deal with weird formats. So each new AI capability is (1) instantly rentable (2) instantly integratable.
In the old API era:
Access. Someone ships an API; developers can finally reach the capability.
Integration (years). Teams wire it into CRMs, billing, support, etc.
Now the second phase collapses. Models read docs, write adapters, transform schemas, even drive headless browsers. The lag from first API call to working in production can drop from years to weeks. When something commoditizes that quickly, value shoots up‑stack to trust, workflow fit, compliance, data, and margins just as fast.
Metaphor: City of locks
Imagine the economy as a city of locked buildings.
APIs are the universal keycards—install a keypad, hand out codes.
AI is a swarm of contractors who (a) use those keycards to go anywhere, and (b) install new keypads on doors that never had them by reading docs, reverse‑engineering locks, and auto‑wiring circuits. Once the doors open, getting in isn’t scarce. Running the building—security, coordination, maintenance—is. That’s the post‑collapse opportunity.
Examples of AI‑Era gates falling (or wobbling)
These are areas where AI is chewing through old gates. Some have dropped; some are shaking.
Translation. Realtime multimodal models (GPT‑4o‑class and peers) eat the permission gate around professional interpretation. Costs for machine interpreting now land at 30‑40% of human interpreter rates in enterprise pilots.
Contact Centers. Voice agents handle thousands of calls in parallel, collapsing coordination + labor gates that used to require large BPOs. Reported cost per automated call ~$0.40 vs $7‑$12 human in some stacks; Replicant & others see ~50% cost/call cuts.
Compute. Groq and other inference clouds rent silicon by the token/second instead of capex clusters; the capital gate around GPU clusters sags.
Short Case Notes
Harvey (law). Sits on top of model APIs. Sells what firms need: clause libraries, accurate redlines, audit trails. Some firms report diligence shrinking from days to minutes. Annualized revenue run rate climbed from ~$50M early 2025 to ~$75M April; adoption across major law firms incl. Allen & Overy pilots with 3,500+ lawyers; reports of top‑10 U.S. firms usage.
Lovable (apps). Natural‑language to app code‑gen. The hard part now is production: CI, deploys, maintenance. That's what they sell. Lovable crossed ARR ~$75M within 7 months of launch from 180K paying users.
Builder playbook
See a wobbling gate? Run this.
Name the gate. What blocks people now—permission, coordination, capital?
Assume an API soon. If a clean interface could exist within 24 months, plan as if it will. What changes the day that happens?
Smash or surf? Being the Gate‑Smasher means living with regulators, capex, uptime. Hard but valuable. Surfing (be a Specialist) means building on the new surface area: trust, data, orchestration
Map the new scarcity. When access is free, what do buyers still pay for?
Build that. Observability, fraud brains, compliance layers, cost guardrails, UX.
Quick checklist
Capture the exhaust (logs, money flows, prompts).
Bolt on trust (security, governance, fraud, compliance).
Polish the experience (dashboards, orchestration, analytics, SLAs).
Seven wobbles zones we’re watching
These are domains where access was gated and a recent AI capability is chewing through that gate. If AI is just “nice to have,” it didn’t make the cut.
Gen‑Video / Virtual Production – Gear + studio time gated; gen‑video lowers bar. Scarcity: taste, story pipelines, rights, rendering budgets.
24/7 AI Tutoring & Curriculum Copilots – Teacher time + district buy‑in gated; AI tutors scale 1:1. Scarcity: proof it works, retention past week two, assessment data schools accept.
AI Clinical Scribes & Chart Copilots – Doctors drown in notes; ambient AI captures visits (Abridge, Ambience, Nabla). Scarcity: provenance, billing accuracy, medico‑legal defensibility.
AI Contract / Deal Desk Automation – Legal queues slow revenue; LLM tools mark up docs fast (Harvey, Spellbook, etc.). Scarcity: audited clause libs, change tracking, risk scoring.
AI App Assembly / Code Agents – Internal tools bottlenecked on engineers; NL→app (Lovable, v0, Cursor) spins them up; scarcity: prod hardening, access/secret guardrails, cost budgets.
Contact Centers 2.0– Scaling support = hire/train armies; realtime agents now handle L1/L2 + CRM writebacks; scarcity: escalation routing, QA evidence, outcome analytics.
AI Robot / Device Fleet Orchestration – Mixed hardware + firmware formats gated ops; cloud/edge AI smooths; scarcity: safety certs, rollback, normalized telemetry, uptime SLAs.
APIs turn yesterday's moat into today's baseline. AI speeds the conversion.
If you’re a founder, find the wobbling gates and build where scarcity is moving. If you’re an investor, ask founders where they sit on the Smash ↔ Surf spectrum.
P.S. Are you building in an active wobble zone? Let’s talk—vedika@weekend.fund.
Thanks to Ryan Hoover for reading drafts of this post.