Elena Burger

Angela Strange traces a16z’s initial international strategy to a line from Nubank’s David Vélez (relayed via her brother): “you could compete to be the fifth financial services provider for every customer in the US,” or go to markets with “5 fat, happy banks that only serve 20% of the population.” That thesis produced her first-ever a16z check—Santiago Suarez’s Addi, now serving a quarter of Colombia’s population for banking and payments—and later intersected with Gabriel Vasquez’s mapping of LatAm’s then-30 unicorns into a WhatsApp community.

Gabriel frames AI’s dichotomy: the technology “is very democratic… it distributes the ability to get it anywhere in the world,” but also concentrates the epicenter of rapid movement in the Bay Area. The flow reversed—instead of a16z flying to Brazil and Colombia, founders everywhere wanted to come to the Bay Area—and country-specific diasporas became the on-ramp. Elena Burger argues these networks may be even more powerful than elite-school alumni networks; Strange says they often lacked organization.

The discussion identifies three native advantages borderless founders can use to accelerate preferential attachment: differentiated talent pools, government-backed brand, and enterprise-customer access. Vasquez’s “AI Olympics” frame: every country wants medal-winners, so Poland’s government invested in ElevenLabs and Sweden’s supported Lovable and Legora, giving them a jump-start in validation with local and adjacent enterprises. Strange adds that “nobody wants to be the first bank or the first insurance company” in the US, but a borderless network can land that first logo abroad faster.

The bridge runs both directions: Cognition’s early go-to-market was Brazil, which represented “a really high share” of its early revenue, because Brazilian enterprises wanted to adopt AI quickly and had fewer providers. The newer, less intuitive power is intra-diaspora cross-pollination—a German company’s first design partner was the largest Spanish conglomerate—which Vasquez presents as a capability that “probably no other investor” can provide from the get-go.

Repeat founders are a deliberate sourcing wedge: a first wave of local entrepreneurs reached meaningful scale, often in the $1–5B range, without seeing their visions through, and “this $5 billion outcome wasn’t enough for me” drives round two. Frederik G. M.’s Pip.com chose a16z despite pre-existing investor relationships after the firm helped ideate with top executives from DoorDash, Lyft, and portfolio companies—differentiation on ideation, not check size.

Practical advice for international founders: visas first—a16z is invested in O-1 visa company Extraordinary—then spend at least 3–6 months in Silicon Valley, not just weeks. Strange says, “I never met somebody that came to Silicon Valley and was like, ‘This was such a waste of time.’” The real product is speed calibration: “Silicon Valley continues to be ahead of every other ecosystem” in the speed at which people operate.

The long-term claim: 40% of a16z’s investments were in international founders, split evenly between those based in the US and those based elsewhere. Vasquez wants the share of venture returns attributed to companies outside Silicon Valley to rise from 10% to 20–30%. He says the statistic does not fully tell the story: many borderless companies are based in Silicon Valley while much of their engineering team may remain in the founders’ home country, so they can be claimed by two countries.

🔗 Original source & video: The New Geography of Startups

Joe Schmidt’s 2x2 offers a framework for evaluating two enterprise AI sales playbooks: lighthouse (high buyer exposure, proof travels—regulated industries and constrained logo sets) and land grab (low exposure, established budget, provable ROI). The compression of the whole framework: “proof on the top right of the quadrant and math on the bottom left” — lighthouse relies on reference logos whose proof travels; land grab relies on showing the buyer the math against whatever human- or software-driven solution they pay for today.

Andy McCall’s Samsara story is a Land Grab example: the 2016–2019 ELD mandate forced the trucking industry to find budget at once, and the new entrant got a boost by selling to the mid-market despite incumbents such as AT&T, Verizon, and players already at “hundreds of millions, half a billion in revenue.” His honest admission: “there wasn’t a lot of strategy… who’s willing to pay us?” — cold calls to the largest trucking and transportation firms as an 18-month-old company got “we’re not buying,” while the mid-market needed less social proof and gave fast product feedback.

The tradeable macro claim: the wedge/PLG era was an artifact of the last cycle, and “there’s a moment right now to go sell big software again.” Joe’s reasoning: the 2000–2008/10 cloud platforms (CRM, HR, ITSM, security) won the platform layer, forcing wedge products and land-and-expand because cloud-to-cloud switching was “green or blue” button indifference — but AI is “not a skeuomorphic, one-to-one replacement,” agents can absorb rote work, and companies can rethink even fundamental platforms.

The portfolio map: land grab = Stuut (AI accounts receivable — “humans plus AI” collections sold on working-capital math to the mid-market) and Pylon (AI-native customer support, climbing the ACV ladder); lighthouse = Harvey (won the first critical law firms and “that proof traveled big time”) and FurtherAI (some of the biggest insurance companies, governance-first, forward-deployed teams). Decagon is the onboarding exemplar: “here are the benchmarks that we are signing up to hit, and then they hit them” in a high-risk, exposed market.

Andy’s ACV discipline: “you think about it a lot and then you try not to think about it at all.” Deals must clear the unit-economics hurdle; past that, stop optimizing — if your engine lives on $15K ACVs, don’t take $8K deals, but grab every $15K one, build a repeatable engine, “pour fuel on the fire,” and inch up the ladder over time.

POC hygiene for the AI era: account for product complexity, then box every trial with a hard end date (30/45/60 days, “period, end of story”) and success criteria defined up front, or it becomes a “science project” — because models improve daily, the answer to “can it also do this?” is often probably yes. Elena adds that when automating never-automated workflows, configuration costs money and “the product works” can be separate from “the product is being used correctly.” Founders should scope explicitly what they are and aren’t signing up for.

The biggest founder mistake is over-strategizing the choice itself: “spend 1% of your time on the strategy… 99% of your time trying to execute,” and “there’s no bonus points for hard-earned revenue” — plus vanity targeting, since it sounds “way sexier” to sell to JPMorgan Chase than to Morgan Chase. Nearly every large company eventually runs both playbooks: Moroi and Samsara both started land grab, then verticalized into lighthouse (school districts, public sector) once mature.

🔗 Original source & video: Choosing Your Sales Strategy: Lighthouse vs. Landgrab

vLLM has become a widely used execution layer connecting open-weight models to major accelerators. Introduced as running on half a million GPUs at any moment, it supports more than 1,000 active model architectures while NVIDIA, AMD, Google, Amazon, Intel, and others ensure new chips can run it—and often benchmark against it. Simon Mo likens its role to “databases and operating systems” for AI.

Open weights shifted from enthusiast territory to strategic infrastructure when application companies needed differentiation beyond a proprietary-model wrapper. Matt Bornstein points to Cursor, Decagon, Harvey, and similar startups requiring their own mid-training, post-training, inference, and deployment techniques. Closed APIs do not provide that access, so open source became “deeply embedded,” even though Matt notes OpenAI and Anthropic models remain more widely used and generally more critical overall.

The economic case is increasingly about controllable performance, reliability, and data—not merely cheaper tokens. A voice-agent company can control its infrastructure and enforce a latency SLA. Kimi K2 bridges almost a 10x price gap without being as expensive as Claude or GPT-5, while bringing an Opus 4.1-level model onto infrastructure that users can run and fine-tune. Open-weight providers can potentially offer 10 speed tiers, including 400–500 tokens per second in some workloads, versus a proprietary provider’s regular and fast modes.

Open-weight licensing is moving away from unconditional gifts because frontier training cannot be sustained by donated developer time. Model labs face millions or billions of dollars of compute plus repeated failed runs, leading to usage thresholds, derivative-work provisions, and commercial agreements. Simon’s pharmaceutical analogy captures the requirement: released products must return enough revenue to fund the next risky R&D cycle.

Moderation failures may make open weights the default for trusted, specialized work. Simon argues that proprietary guardrails remain arbitrary and false-positive-prone; even GPU-kernel debugging can trigger restrictions and destroy a two-hour session. “If moderation is never solved,” users will prefer models whose guardrails they can control for trusted use cases.

Simon expects no meaningful open-versus-closed capability gap within one year because progress now depends more on environments and algorithms than distribution strategy. Moonshot’s front-end coding loop—generate, render, inspect, and iterate—is his key example of an environment that cannot simply be distilled. He leans against distillation as the main explanation for progress: the durable engine is “really smart people” combining compute, data, environments, and novel methods.

🔗 Original source & video: How Open Source Became AI’s Backbone | Inferact with a16z

The UI is losing its monopoly on access, but the system of record remains valuable. Seema Amble calls Salesforce’s Headless 360 largely a rebrand of existing APIs, yet an important acknowledgment that agents may retrieve CRM data without opening Salesforce. The durable asset remains “the data, the logic, everything stored below it.”

Incumbent enterprise software is protected less by screens than by decades of encoded business logic and exceptions. A PostgreSQL database plus APIs cannot simply replace SAP: deployments codify how a 100,000-person, 20-country company operates, complies, and decides. Steven Sinofsky’s blunt warning is that founders “wildly underestimate” the sophistication customers have built into these systems.

“Agent” obscures three economically different jobs: lookup, action, and analysis. Lookup is mostly a more forgiving interface; action raises identity, permission, credential, and paid-seat questions; analysis can span systems and models but requires verification because hallucination becomes consequential. Headless access therefore does not by itself solve enterprise deployment.

The long tail of exceptions—not the routine workflow—is the central agent challenge and a potential source of new product value. Geographic practices, account-specific judgment, permissions, and policies often live in employees’ heads rather than CRM fields. Agents can collect that context by observing calls and computer use, but “almost everything interesting in an enterprise is an exception,” so trust accumulates slowly.

Automation is more likely to expand enterprise software demand than finish a fixed quantity of work. Amazon’s automated returns created a new optimization loop; automating expense and travel workflows can produce new performance analysis; AI-assisted contracts may become longer and more sophisticated. Sinofsky’s framing: “The long tail got no shorter. It just got longer in a different way.”

The strongest startup wedge is between established categories or organizational functions, not directly against a mature incumbent. A head-on replacement inherits “8,000 things” a customer expects, while an AI-native overlay can translate between sales and finance, convert collected data into action, or capture previously invisible field activity. Sinofsky’s instruction is simple: “Aim for the middle and do things in the new way.”

Enterprise AI’s most credible network effects may form inside companies rather than across them. Compliance and security make external networks difficult, but visible wins with chat can spread among colleagues, much as advanced Excel use once did. Products connecting functions that previously needed manual integration could create entirely new categories.

🔗 Original source & video: The New Rules of Enterprise Software with Steven Sinofsky