Why I work with them

Decide fast, are frontal about why. First call to first answer is days, not months. If they’re in, you know. If they’re out, you know that too — with the actual reason, not the LP-friendly version. That clarity is rare and worth a multiple on its own when you’re running on a finite cash runway.

Best-in-class on fundraising. They’ve taken more LatAm companies through their next round than almost anyone else. If your strength is building product and the next round terrifies you, this is the fund that complements you. They’ll prep the deck, the narrative, the warm intros, and they’ll do it without making you feel like you owe them a favor.

The Brazil door, opened. I’ve watched founders try to crack BR for years from the outside. Canary is the network that gets you the first 20 customers, the first key hire, and the first credible local press hit, fast. If you’re a non-BR founder building for the region, this is the launchpad I’d want on the cap table.

The bet

For founder-builders who’d rather be coding than pitching, the partner you want is one that’s clear, frontal, and exceptional at the part of the job you find painful. Canary is that partner.

The demographics are inverting. People are living longer, families are smaller, and there are fewer young people around to take care of the old ones — in every country I look at. Loneliness, missed medications, no one to read them the news. The gap between what seniors need and who’s there to give it to them is widening every year, and it’s not going back.

Ato is a screen-free voice device that sits in that gap. No smartphone, no apps, no learning curve — just talk to it. It listens, it answers, it reminds them, it calls their family. The kind of thing nobody was building for them, even though it’s the most obvious thing in the world to build.

Why I backed it

Two founders, very young, very high energy. I’ve known Juan for a long time and watched him build many different things. He’s a programmer, but more importantly he’s a crafter — he does whatever the build needs, in whatever discipline. That kind of operator is rare and disproportionately valuable at the zero-to-one stage when no role is yet defined.

They went straight to San Francisco. Juan and Gaspar packed up and moved early — cracks who chose to put themselves inside the conversation that matters for hardware-plus-AI, before the company demanded it. That takes guts and it compounds.

The product is named after Juan’s grandfather. He was Ato’s first user. That’s not branding theater. It’s a forcing function — when the company is literally named for the person you’re building for, you can’t lie to yourself about whether seniors love the thing. Things built in honor of someone the founder loves tend to be better than things built off a deck.

The right tier. This is tier-2-leaning-tier-1: the device proposes the action (call your daughter, time for your pill, want to hear today’s news?) — not just answers when poked. That’s the right shape of AI product for this user.

The bet

Voice-first hardware for seniors is one of the cleanest demand pulls in consumer AI. The user cohort is enormous, undertargeted, and willing to pay. Ato can become the default device in that category.

Why I backed it

Andrés. I’ve been helping him with intros to VCs and angels and advising for a while now, and what he’s built as a solo founder is genuinely impressive — 4,000+ developers organic, full multi-tenant platform, observability, flows, the works. His AI-first bias is the right one for this moment, his founder style is the kind I want to keep working with for a decade, and the references I have from mutual friends are unanimous. I offered him a small angel check.

The vibe-coding stack tailwind. Lovable, Claude, Codex, Cursor — these are becoming the default way internal software gets built across LatAm. Every internal tool that gets vibe-coded into existence eventually wants to talk to a customer, and in this region that means WhatsApp. Kapso is positioned to be the native, drop-in infrastructure layer for every one of those apps. The buyer here is the developer, not the IT director — and the developer is increasingly an AI.

Personal autonomous agents in WhatsApp-first markets. In regions where WhatsApp is the economy’s OS — most of LatAm, India, much of Southeast Asia — autonomous agents (Claude Code, Open Claude, the next generation of personal agents) need an integration layer to act on a user’s behalf inside WhatsApp. Kapso is the obvious place that layer should live. There’s no incumbent here yet.

Tracked the space since Agentmail. I tried to get into Agentmail’s YC round and couldn’t — close-but-no-cigar. The fact that Kapso is the next move in roughly the same vector, with a founder I already know and a region I understand better than email, made the bet much easier the second time around.

The bet

Two stacking tailwinds — vibe-coded apps + WhatsApp-first agent users — both routing through the same infrastructure problem, both addressed by the same product, in regions where the buyer is technical and willing to pay. If Kapso becomes the default import * from kapso for any app that needs a WhatsApp surface, the company prints. Andrés has 4k organic devs already with no marketing engine; the wedge is sharp.

They re-type all of it into the ERP.

Galo eats the mess on intake AND builds a smart-CRM data layer on top: which clients buy what, when, how often, what they stopped buying, what they could be cross-sold.

From the data, it triggers automated actions that drive more sales. The platform isn’t just order automation — it’s the sales engine sitting on top.

Started in food, where Tomás’s network and depth gave the wedge its first cut. Expanding the aperture across retail B2B — household goods, pharma, hardware, anywhere a distributor lives on WhatsApp.

Why I backed it

I worked with Benjamín at Darwin first. He was on the early team — relentless, fast, the kind of operator who doesn’t wait for permission. When he left to build Galo I knew the bet was on the founder before it was on the market. Young, hungry, real entrepreneurial energy. That pattern compounds.

Tomás brings the market. He built SimplEat and grew up in family gastronomy businesses — he knows distributors, restaurants, suppliers from the inside. Founder-market fit isn’t a vibe-check here; it’s deep in the bones. The Benjamín × Tomás combo is the reason this company will know the customer before its competitors do.

Vertical only works if you integrate both sides. I don’t have a generic preference for vertical — Darwin is horizontal on purpose.

What I respect about a vertical bet is the discipline it demands: you need deep industry knowledge AND you have to integrate both the external surface (CRM, sales, customer comms) and the internal stack (ERP, inventory, billing).

Horizontal players touch one side — pure CRM, pure communication, pure internal automation — because they have to be portable across industries. A vertical company that only does one side is just a worse horizontal product.

Galo went vertical AND committed to both internal + external. That’s the move that turns vertical into a real moat instead of a smaller TAM.

The data layer is the moat. Once Galo is processing every order for a distributor, it knows the customer base in ways the distributor itself doesn’t — buying frequency, basket-size drift, churn signals, latent cross-sell opportunities. That data is the asset, not the order automation. Order intake is the wedge; the data + automated actions on top is the durable business.

Real ROI in pesos. A distributor can quantify this: “we saved 4 hours of admin per day per branch, and lifted basket size by 9%.” Two numbers on the cover slide of every renewal. Easy to sell, hard to churn.

Right bet on Build for a World Where Every Customer Gets Their Own Software. Galo isn’t building a generic order entry app — it’s building one that adapts to each distributor’s catalog, pricing, ERP, and accent. That’s the right shape of B2B software for this cycle.

The bet

Retail B2B distribution is one of the largest fragmented and informal industries in LatAm.

The platform that becomes the AI sales layer — order intake + data + automated actions — between WhatsApp and ERP for that segment is going to print money.

Food is the starting vertical because that’s where the founders have an unfair advantage. The playbook generalizes across retail.

Early, but the wedge is sharp and the moat (data, integrations, switching cost) compounds with every customer added.

Why I backed it

The deer market, picked on purpose. Mercately is selling to mid-SMB merchants — exactly the animal that gives you optionality. They can move down to mom-and-pop self-serve later, or up to enterprise integrations. Right now the deer is paying.

LatAm-native channel. This isn’t a US playbook ported to Spanish. WhatsApp commerce is a different physics from email/web checkout, and the team built for the local reality from day one — Stripe + HubSpot integrations, MIA the AI agent, multi-country compliance. The playbook in São Paulo doesn’t ship clean to Boise.

Profitability before the round. The company hit $1.5M+ ARR before raising the $2.6M seed. That’s a founder discipline I respect. They didn’t burn capital looking for product-market fit — they sold first, raised after.

The bet

Conversational commerce is a generational shift in LatAm. The platform that owns the rails between WhatsApp + payments + AI for mid-market merchants is going to be a real outcome. Mercately got there first with discipline.

Why I work with them

Operator-led. Brian was an operator before he was an investor. That changes everything about how the firm shows up — the questions, the introductions, the help on hiring, the help on positioning. Tomi is genuinely founder-oriented, in a way that you feel within the first call, not on slide 12.

They actually help. Intros that close, hiring referrals that match, a network in the US that LatAm founders usually have to spend years building from scratch. That bridge — LatAm operating reality on one side, US capital and customers on the other — is the thing they’ve built and it’s hard to replicate.

No later-stage conflict. Latitud invests at pre-seed only. Most multi-stage funds will say they’re founder-friendly, but the moment your Series A starts moving, their incentives split — they want to lead, set terms, take board, signal to LPs. A pure pre-seed fund just wants you to win the next round at a higher price. Their next dollar isn’t competing with yours.

That’s why they feel more like a co-founder than an investor. They show up early, help the most when it matters, and step aside when the bigger checks arrive.

The bet

Founders in LatAm are still capital-efficient by necessity, and they need the people on their cap table to add disproportionate value per dollar. Latitud is the firm I’d put first on that spec.

Why I backed it

Right channel, right buyer. Same instinct that worked at Sirena: meet people in WhatsApp, not in a portal nobody opens. Lara understood that employee experience tools fail because employees don’t go to them — so they went to where employees already were.

Founders who’d been operators. They’d built and sold inside HR teams before founding the company. They knew what HR leaders ignore vs. what they actually pay for.

Strategic exit makes sense. Visma has 4.5M users across LatAm HR tech. Lara plugs into that distribution overnight rather than burning years building a sales team. The founders chose the right buyer, not just the highest one.

The exit

Acquired by Visma in April 2025 (terms undisclosed). Clean strategic outcome: Visma needed an AI HR layer for the LatAm portfolio; Lara needed enterprise distribution. The kind of acquisition where both sides win — which is rarer than the headlines suggest.

Why I backed it

Boring industry, monstrous TAM. Every bank, every fintech, every BNPL provider in LatAm has a collections problem. The buyer is the CFO, not the CTO. Sales motion is straightforward: “we recovered 25% more for [reference customer]”. Numbers do the work.

Compliance as moat, country by country. Kleva is compliant in Mexico, Brazil, Colombia, Guatemala, Ecuador, Peru — each with different regulations on contact frequency, scripts, escalation. That regulatory layer is exactly the kind of moat I think actually holds in AI in LatAm. Every country adds switching cost.

The right kind of replacement. This is inhuman work — repetitive, scripted, emotionally draining. Replacing it with AI is the canonical “let humans do human things” move. The few humans left in the loop become managers and exception-handlers, which is the better job.

The bet

LatAm financial services is a sleeping giant for AI applications. Kleva is in the highest-leverage workflow with the cleanest unit economics. If they keep their cost-per-recovery curve heading down, they become infrastructure.

Why I backed it

Whole-job replacement, not feature add-on. Samu isn’t a copilot for sales reps. They’re automating the work end-to-end — record, transcribe, score, update CRM, surface coaching. Tier 2 minimum, often tier 1.

Both founders worked with me at Sirena. Andrés Bruzzoni built the entire Inside Sales / SDR team there from zero — outstanding operator, and the heaviest user of every call-coaching tool we ever rolled out. He lived the pain in his own seat for years before deciding to build the answer. Jonathan Sosin was on the engineering side at Sirena — very, very strong technically. Pattern I trust beyond pattern: founders who used the thing they’re now building, in the same region, before starting.

LatAm sales doesn’t only happen on Zoom. US-built tools (Gong, Chorus) optimize for video calls because that’s the US sales motion. LatAm sales is messy across surfaces — WhatsApp threads, voice notes, calls on personal phones, plus a lot more presencial in physical stores. The team that ingests all of those and turns them into one coherent CRM signal wins. The team that only ingests Zoom/Meet is porting a US product into a market that doesn’t sell that way. Samu chose to sit across the surfaces — that’s the part the US-first players will struggle to copy.

The bet

Sales coaching cuts ramp time for new reps from 6 months to 6 weeks when it works. Across LatAm SMB sales teams, that’s a number every revops leader will pay for. There will be one or two dominant sales-intelligence platforms per region. In Spanish-speaking LatAm + Southern Europe, Samu is positioned to be it. 500 Global already saw the pattern.

Why I work with them

The Switzerland of LatAm VC. Norte writes follow checks only. They never lead, never push terms, never compete with another fund on a board seat. That stance is the source of their power — every other VC, every founder, every operator can talk to them without strategic friction. The result is the densest network of relationships in the region, and they get into the rounds that matter because everyone wants them on the cap table.

A working index of LatAm tier-1. Their portfolio reads like a map of who’s actually building the next generation of LatAm companies. If you want to understand the regional cycle, look at where Norte is putting follow-on capital. The signal is cleaner than any tracker because they only get in the rooms that already attracted tier-1 lead investors and tier-1 founders.

Best events in the region. Gustavo and the team run events you actually want to be at — the connections, the calibration, the chance to compare notes with operators a year ahead of you. For a market as scattered as LatAm, that physical convening matters more than people realize.

Young partners with gen-Z reach. Most LatAm VC firms have a partner-age problem — they hired in 2014 and never refreshed. Norte’s partner bench is genuinely young, which means a real connection to founders building right now, in 2026, on the platforms gen-Z founders default to. That’s not cosmetic; it’s how they keep finding the next cohort early.

The bet

LatAm needs a Switzerland-grade follow fund — neutral, deeply connected, conflict-free — to grease the rounds that get the best companies funded. Norte is that fund. If they’re already in your round, you’re probably in the right round.

Why I backed it

A new buyer, ignored. Most “AI for sales” tools are sold to revops at 200-person SaaS companies. Creators are an asymmetric segment — they have real revenue, no infrastructure, and almost no tools that respect their workflow. The pricing, ICP, and product surface are different enough that a focused team wins.

Ride the wind, don’t bet against it. Creator monetization is one of the strongest gusts in the AI cycle. AI agents lower the cost of “having a sales motion” by an order of magnitude — exactly the kind of compounding capability that lets a one-person business hit numbers a five-person team used to need.

A second-time founder leading a young, content-native team. Daniel built and sold a company before — he’s done the painful loops. The team around him ships fast and creates its own content, growing audiences while building the product creators need. The pain is theirs, not interviewed.

End-to-end, not horizontal. From where I sit at Darwin, creators are one of the most under-served buyers in tech. Horizontal tools — generic CRMs, AI-SDR-of-the-week — help around the edges, but they don’t give a creator an end-to-end loop. Ninjō can own the whole thing: marketing → conversion → ROI in a single product built for one-person businesses that do real revenue. Much more interesting than another horizontal sales tool fighting for revops slack-channel attention.

The bet

Creators-as-businesses is an unbundling of the creator economy. Whoever builds the AI sales infrastructure for this cohort becomes default — and Ninjō is positioning early.

Why I backed it

Tool for a clear archetype. Volt is built for someone with a phone full of clients, projects, and chats — founders, account managers, freelancers, recruiters. The kind of user who’d pay $20/month to save four hours a week without thinking about it. That’s a clean willingness-to-pay.

Privacy as moat. Local processing. No Volt server sees your chats. Zero data retention with AI providers. In a world where every WhatsApp tool is a vector for data leakage, “we literally cannot read your messages” is a real positioning win — especially for the audience that has confidential conversations going through WhatsApp daily.

Productivity tools have margin. Selling to professionals who measure their time in money. Different unit economics from selling to teenagers or SMBs.

The bet

WhatsApp keeps eating workflow tools (calendar, CRM, Slack-replacement) in LatAm + Southern Europe. The power-user layer is a real category, not a feature. Volt is positioned to own it.

Why I backed it

To fill in.

The exit

To fill in.

Why I backed it

Untapped data layer. This is one of the largest unmonitored data surfaces in retail. Hours of customer conversations every day, none of it captured. Vera is doing for in-store what Gong did for sales calls — and the analogy holds.

AI as gravity field, prompts as masses. This is a tier-1 use of AI — not “transcribe the meeting”, but “tell me what conversion-killing behavior the team doesn’t know it has”. The model proposes; the operator decides what to fix.

Founders who shipped before raising. Small round ($600K seed), real customers, narrow focus. They aren’t pretending to be the next Salesforce — they’re picking a wedge and going deep.

The bet

Brick-and-mortar retail still does ~85% of US consumer spend and a higher share in LatAm. Whoever turns the in-store conversation into structured signal becomes infrastructure. Early and small, but the wedge is clean.

Why I backed it

Whole-job replacement, not feature add-on. Selenios isn’t selling “AI for recruiters” as a copilot. They’re automating the task end-to-end — sourcing, scheduling, screening, scoring. That’s tier-2 minimum, often tier-1. The right shape of AI product for this cycle.

Founders out of Silicon Valley, building in Buenos Aires. Esteban Zecler, Julián Bender, Jonathan Muszkat — they left big-co salaries to build in LatAm. Pattern I trust: people who’ve seen scaled SaaS up close, choosing to build the next thing in the region they understand.

Real numbers fast. 35+ customers in four months, $1.2M pre-seed, multi-sector traction. Not a deck — a product people are paying for already.

The bet

The cost of hiring a candidate drops 5–10x with end-to-end agentic recruiting. Whoever owns that workflow in LatAm before the multinationals do is going to be a default purchase for every mid-market HR team.

   H+  H+  H+  H+  H+        · gradient, on loan from the sun
  ═══════════╦═══════════
             ║  ◷           · the motor spins, ~100×/sec
  ═══════════╩═══════════
        ADP + Pᵢ  →  ATP    · order, briefly, locally

Everything in the universe rolls one way: hot to cold, ordered to scattered, structure to dust. That’s entropy, and the second law of thermodynamics has no exceptions — except one strange, stubborn pattern. Life.

Watch Destin Sandlin take apart ATP synthase — a real rotary motor, smaller than a virus, spinning hundreds of times a second inside every one of your cells right now. It builds order: it manufactures ATP, the molecule that powers you, one assembled rung at a time. A motor. Made of protein. Billions per cell.

But it cheats. The motor only spins because protons pile up on one side of a membrane — a gradient, borrowed, ultimately, from the sun. Life doesn’t break the second law; it surfs it. It builds a pocket of order here by dumping a bigger mess there. Net entropy still rises. Always.

Naval put it best:

Humans locally reverse entropy because we have action. In the process, we globally accelerate entropy until the heat death of the universe.

So you’re not really a thing. You’re an eddy — a standing whirlpool in a river that only flows downhill. Structure on loan, paid for in disorder somewhere else. The bill always comes due. But the spinning, for now, is the whole point.

Sources: Smarter Every Day #300 — Nature's Incredible Rotating Motor; the entropy line is from The Almanack of Naval Ravikant.

                                         ...:::::-----------::::....                                                                        
                               ..:--=++***######################*#****+++==--::.                                                            
                        .::-=+**#######********+++++++++++++++++**********#####**+=-:.                                                      
                    .:=**######*****++++++++*+++*++++*+*++++++++++++++++++++++***####*+-.                                                   
                  :+#%%##******++*+++*+++*+**+****+**+++*++****+**++**+*******++++****##*=.                                                 
                .+#%##****+++**+***+**+************************************++++**********#*=                                                
               .*%##****+*+*+*+**+*+*****+********+*+**+*++++++++++++++++++++++++++********#=                                               
              .*##*****++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*******#=                                              
              -##****++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++******#:                                             
              +#**#**++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++******#-                                             
             .+*****+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++******%=                                             
              +#****+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++******#-                                             
              -#*****+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++****#*##.                                             
               +******++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++****#*#*                                              
               :#*****+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*******#-                                              
                -#*****++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++***+*****#:                                              
                 =#*#***++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*********#+.                                              
                  =#*#***++++++++++++++++++++++++++++++++++++++++++++++++++++++++**********#=                                               
                   +#*#****++++++++++++++++++++++++++++++++++++++++++++++++++++++*********##:                                               
                   .+**#****+++++++++++++++++++++++++++++++++++++++++++++++++++******+****#+.                                               
                    .***#****+++++++++++++++++++++++++++++++++++++++++++++++++************#+                                                
                     .***#*****+++++++++++++++++++++++++++++++++++++++++++++*************##-                                                
                      .***#*****++++++++++++++++++++++++*+***++++++++++++++*************##*.      .-==++++*********+++===--::.              
                       :***#*******++++++++++++++++****+*++++****+++++++++**+***+********#+      -%%%%%%%%%%%%%%%%%%%@@@@@%%%#*=.           
                        :**********++++++++++++++++*+***+**++*+**+++++++*+**+************#=      +@%%%%%%%%%%%%%###%%%%%%%%%%%%%#-          
                :=+****#*#%######****+++++++++++++****+***+++*+*+++++++*****************##-      .+#@@%%%%%%%%%%%@%%%%%%%%%%%%%#%%.         
               =%%%%%%%%%%%%%%%%%%%%%%#**++++++++++***********+*+++*+*****++++************:        .=*#%%%%%%%%%%+*%@@%%%%%%%%%@@+          
               *@%#######%%%%%%%%%%%%%%%%#*++++++++*********************++++++**********#*.          .*%%%%=::::.  .-*##%%%%##*+:           
               :*#%%%%%%##%%%%%%%%%%%%%%%%#***+++++*******************++++++++**********#*.          :###%=             ....                
                 :++###%%%%%%%%%##%%%%%%%%#******+*****************************######*#*#*.          +##%*                                  
                    -#%%=:-+%@%%%%%%%%%%#***#**#**************#****************########*#*         .+###%-                                  
                    -#%+   -######%%##*******######*##########****************#######%%#%*========+#%##%*                                   
                    +*%-  -#*****#%%#**********##############*****************#######%#%#%%%##%%%%%%%%%#.                                   
                   .**%-.:*%##%%%%%%##**********###############*##*#********########%%%%%%%%%###%%%%%%#.                                    
                   +#*%######%%##%%%##***********###############*#*#*********#######%%%%%%%%%#%%%%%%%+.                                     
                  :#*#%##########%%%#*#**********##################*###############%%%%%%%%%%%%%%%#=:                                       
                  -#*#%%%%######%%%%###****#***#*##################################%%%%#%%@#---::.                                          
                  -##%%%%%######%%%%##*#######################################%####%%%%.-@#:                                                
                  :#@%####%#####%%%%#######################%#######################%%%#. .            ..--==--:.                            
                   :-=%@#:*%#%%#%%%%###############%#########################%%###%%%%#            .-+*###%####*+-.                         
                     .+*- +%#%%%%%%%#############%#%%%%######%###############%####%%#@*         :-+################+-.                      
                          +%*#######*###############%%%#%###%######################%#@+     .-=*#####################*=.                    
                          +#**#%%#%##****###########%#%#%####################%#####%%%=..:=+##########################%#-                   
                         :%%####@#%%%%%##**#######%%%%%#%###############*****#***********###**********************######%+                  
                          +@%###%#%%%%%%%%##**###########***********************************************************######=                 
                          +%####%%%%%%%%######**#************************************************++++++++***********####*#*                 
                          +#*#*%%%@%%%%%%%%%%###########***********************************+++++++++++++**********###*****:                 
                          =###*#%%@%@@%@@@@@@%%%%%%##*#########*************************++++++++++++++*********#######*+-.                  
                          -##%##%%%@@@@@@@@@@@@@@@@@%%################******************************#*###########**+=-.                     
                          .+###%#%%%%@@@@@@@@@@@@@@@@@@@@%%%###*******####################################***+=-:..                         
                            =**%%%#%%%%@@@@@@@@@@@@@@@@@@@@@@@%%%%%##***++********#####**********#*+++=--:..                                
                             :+*%%%%%%#%%%%%%%%%%%%%%@#*#***#%%%%@=:..       ......:::=*#%%%@@%%%@+                                         
                                -+#%%%%%#########%###%#-    =@%@%@+.:::-----========--::::-=++#@%@+                                         
                                  .:=*##%%@@@@@@@%%%###%**###@@@@@@@@@@@@@@@@@@@@@@@@@@@%#+   =@%%=                                         
                                       .:::---===#######%@@@@@@@@@@@@@@@@@@@@@%%%%%%@@@@@@@#=:=@@%-                                         
                                                 -#########%@@%@@@@@@@@@@@@@@@@@@@@@@@@@%%%@@@%@@%:                                         
                                                  +###*##%#%@@%%%@@@@@@@@@@@@@@@%@@@@@@%@@@%%%%%%@=         .:::..                          
                                                  :*#####%##%@@%%%%%%%%@@%%@@@@%%%%%%%%%%%%%%%%%##%=       :#@%%#**+++=.                    
                                                   +%%###@%@@@@%%%%%%##%%#%#########################****###%%#%#%%%@@%@:                    
                                              .:::+%%%%%#@@@@@@%%%%%%%#***************################*+++==#%%%%#*=:::                     
                                             -#@@@@%%%%%%##########%%%#***************######%%%%%%%=        .===:                           
                                            .#%########%%%%%%%%###%%%%##########%%%%%%%%##**+++===-                                         
                                             -#%%%%%%##=--=--::.::-=+*#+**===---==---:::.                                                   
                                              ........               :-:--                                                                  
                                                                    .:-:-=                                                                  
                                                                     :-:--                                                                  
                                                                    .:-::-                                                                  
                                                                     :-.:-                                                                  
                                                                  .:--=:-+:.                                                                
                                                                 .*%%###*#@%:                                                               
                                                                 .*#*##***%@-                                                               
                                                         ....::--=*#####*#%%+--:::...                                                       
                                        ....::---==++++**###%%%@@%%%%%%%%%%%%@@%%%%%##***+++==---::...                                      
                                    .=+**##########%%%%%%%%%%%%%%%%%%%%%%%%##%%%%%%%%%%%%%%%%%%%%%%###**=.                                  
                                    +##############%%%%######%%%%%%##%%%%%%*%%%%%%%%%%%%%%%%%%%%%%%%%%%#%+                                  
                                   .*#######%%%%#############%%%%%%##%%%%%%*#%%%%%%%%#######%%%%%%%%%%%###                                  
                               ..=+*##%%######******###########%%%###%%%%%%*%%%%%%##################%%####*=-:..                            
                         .:-=+*#######***************#################%%%%%##%%%%%%%%%%###################%%%%##*+=-:.                      
                  .:-=+**#%#%##****************#####****++==--::..*###%%%#%%-.::--==++**####%%%%#################%%%%#*++=::.               
              .-+*####%####****************+++==--::...           +#######%%:           ..::--==+############################*=:            
             .+###**######*****++*##%###%*:...                    +#######%%.                ...:+#%@###%+==+++**#########**#**#-           
             .***+*%%*==--:::....-=++=+++-:.............          +#*#####%#. ..............::::-=+++====:      ...::--==*#***+#+           
           :-+###*##:           .::::::::-........................=#*#####%#..................  ...........            .-+###**#-           
          -**%@@#*#%:           .::::::::::..................::::::-**####*-:::::............   ..........            -**#%@@#*#+           
          ++**###*##+          ..::::::::::......:::::::::::::::::::**#***+::::...........................            =***##%####:          
          +**#**%#%#*     ........:::::::::::::::::::::::-::--------****+#*:::::::........................            -***#*#%%%%-          
          :**#%#**#@=.:.............:::::::::::::::::::::::::::::-=+****+#+:::::::........................          ..:*##%@#*#%*           
         ..=*%%**##+-:::........................................-+*#@@%#*%+..............                        ....::=*##****+.           
           ..:::::::.                       ....................+***#%%**##: .                                         .......              
                                                                +**#**#%%##-                                                                
                                                                -***#%#*#%%:                                                                
                                                              ..:+#%@%###%-..                                                               
                                                                ..:--:--:.

My neck hurt for years. Bad posture — slouched at a laptop, hunched on a pillow at night.

Three things fixed it:

1. Sleep posture. One pillow, neutral spine. Stopped curling into the side.

2. Daily exercises. Neck and upper-back mobility, plus McKenzie press-ups for the worst weeks.

3. The Herman Miller Aeron.

The Aeron isn’t just comfortable. It actively corrects you. The back contour and the lumbar support make slouching feel wrong — you sit up because the chair won’t let you not. After a few weeks, my default posture changed even when I wasn’t sitting in it.

$ a tiny history

Herman Miller started as a furniture shop in Zeeland, Michigan, in 1905. By mid-century the company was home to Eames, Nelson, Noguchi — quietly responsible for half the chairs you’ve seen in design museums.

The Aeron launched in 1994, designed by Bill Stumpf and Don Chadwick. People hated the look at first — a skeletal mesh thing with no foam, no leather, no warmth. Then the dot-com boom hit, every startup bought twenty, and the chair became shorthand for “we’re hiring.” When the bubble burst, eBay flooded with them.

It was the first chair ever inducted into MoMA’s permanent collection. Thirty-plus years later, still in production.

Worth every dollar.

......................................... -#........................................................
.........................................:@* .......................................................
........................................ +*#: .......  .............................................
........................................ =#:#=. ....... ............................................
...................................  .... #-.+#: .....+- ...........................................
.....................................  ...=# .-%-.... =@:...........................................
....................................+*... #+..:=# .  :#@- ..........................................
................................... -@* .**.. :-%  :+*=#.........  .................................
..................................  #+#-#*:... *+ -#= +* .  .....=. ................................
................................. .#+.%%=:... -%..%=:.=%  *: ....*%.................................
..........................  .... -#= -@=:.....#+ .%:.. +#=@#...  *@+ ...............................
...........................: .. =%-..:@-.   .:#@-=@- .. -%:+*. .**=# ...  ..........................
...........................*+...%-:.. -:..:..--#@@*... .   -% =#= %- .. -: .........................
.......................... =@-..%-:.    ..-#..:::. ..=: ..:**=#- +# . .+% ..........................
...........................##*  %-:.=.... =@= .....  *# .:-@=%-:.#+ .- #%: .........................
..........................#=.%**#.. *+ . =##=..  . .=@+ .:=@@@...-%:-@-:*#:.........................
........................ -%...==.  :@= .**:%..... =+#+...:-*#=..  -**=% .@-.........................
.........................:@-.   ..-##..#+..%: .. =+=+ ..  ..  .-+....=%.:+.  .......................
.......................   +@+:....#:#..%.- :+:.. +:*=   +- ... -@-..:%: .  :........................
.......................*####*=.   ==-+=%+:.  .....-:*=-=##....-++:  -@++..=% .......................
..................... :%----==++==-==-::--.     .... ::::.  .:----===#@+ .%@- ......................
......................++..    ..:--==+++===--:..   ...::-==++++++==-:%-.: #=#- .....................
..................... =*..   ..       ..::-==+**++++***++==-::..   .:% .%+%::%:.....................
......................:%+-:...      ...      ===#*==--:::...........:%:.##-.:*+.....................
.....................#@#*+**+++=--:.        :+==%--::::..  ....::--==*%*-:...%= ....................
......................=#-==+=***####**+=-::..===%-:::::--==++++++++++++*%@=.*#... ..................
.....................  %:   +* .::-=+**#%%%#*##*##+++++++======+++++==-:%-.#+  =#...................
.................. +* :%.  :%#:        ...:-==+**###*+--=++++++=-:..-  :% -@::=#@: .................
.................. ##*%*--:=*:#-      ...          .++%*+=--::...  .@= .#-.=**=:%...................
.................. *+.--#@%@= :@==-:..        .....-+=#::::::.... :**#=+*@+.. .%= ..................
..................  #+ -#.*+ .*#######*+==-::...  .-=+*::::::--==*#:.%*++*@%.=%- .. ................
..................+= %-*=:* .*= ..::-=++*#####**+==-*=%+++***++++%*  +#=-=#.-%. =*..................
................. =@*%-#==+ :#...        .::-=++*##*#*#*++++++++=-#:.#:..*+ #%.=@#..................
................. ++-- -#=#..*:    ......      ..*-:-**#+++=--:..:#:.=...=# :#%#:+* ................
..................%: .=-=+*+=**=-::.      ..... =*  ==%=-:::.....-.   .:-=##- .  +# ................
..................%- :@+=+++++++**+++=--:..     ++..++#::::::....:-==++++++*%%-.+%:.................
................. =%- -* ..::--===++*******+=-:.-*. +=#::::--=+++++++++++++=%--#+. .................
.................. -#++# .       ....::-==+*#####%*+*+%+++++++===++++=-::..=# @#:=..................
................... :*@=..       ..        ..::-=+*##%%%+===+++++=-:.......-%.-*+-..................
.................. +#*#%###*=-:..       ....       ..:=+*%++=-::..........:-##=:: ..................
................. :%. :+=*=*#%%%#*+=--:..       ..... +:#=::::::....:-==++**+*#@#...................
..................+* .+:+.   .:-=+**####*+=--::...   .+:%:-::::-==++********+=#+ ...................
................. =#  +:* ...      ..:-=++**#***++==-:*-#+==++*******++==-:.  %:....................
..................:%+-+-+     .....       ..:-=+*****+*+*#******++=-::.   ...:%:   .................
.................::=*###%*=-:.      ......     .+-*-==+*++**++=-::............+#+:..................
................  ....:-=*####*+=-:.      .....-=+:   ++%+=--:::.........:-=+**##-:.................
....................    ...:-=+*####*+-:..     =-+...:+#+::::::.....:-=+*###*+-:... ................
........................     ....:-=+*####*+=-.=-+.  -+#=:::::::-=+*###*+=-:.     ..................
..............................    ....::-=+*######=:.:=*+:--=+**##*+=-:...   .......................
...................................     ....::-=+*###*##@**###+=-:..     ...........................
.........................................      . ..:--=+++=-:..     ................................

Censorship used to need a censor. Someone burning a book, someone rewriting a textbook, someone calling the printer.

It doesn’t anymore. In the AI-mediated web, disappearing from the index is enough.

the model is the new front page

For twenty years, reading the web meant opening a page. You typed a query, you got ten blue links, you clicked one.

Now you ask a model. The model reads for you and tells you what’s there. The page itself is something you almost never see. People talk to ChatGPT, Claude, Perplexity and Gemini more than they talk to the source of the information. The author who wrote it, the site that hosts it, the journalist who reported it — all sit one layer below a chat window most readers never leave.

This sounds like a UX shift. It isn’t. It’s a substrate shift. Once a page isn’t in the training set or the retrieval index, it stops existing for the median reader. Not “harder to find” — gone. The reader doesn’t know what they didn’t ask, and didn’t get told.

The same thing happened to card catalogs, to phone books, to RSS. Each layer wasn’t deleted — it just stopped being the layer anyone consulted. After a while, “no longer consulted” becomes “no longer real.”

AI passed human writing in November 2024

This wouldn’t matter much if the underlying corpus stayed representative of human writing. It isn’t.

Graphite scored 65,000 English-language articles from Common Crawl, sampled monthly from January 2020 to May 2025, with Surfer’s AI detector. AI-generated articles crossed human-written articles in November 2024 and reached 51.7% by May 2025.[^1]

A caveat worth keeping honest: only about 14% of articles ranking in Google Search are AI-generated, per the same study.[^1] Publication volume is not the same as visibility. But training corpora are built from publication volume, not from search rankings. The model sees the slop. The reader sees the curated front.

The web is not getting smaller. It’s getting thinner. More tokens, less signal.

the model is training on itself

When a model is trained mostly on output from earlier models, the rare tails of the original distribution disappear first.

Shumailov et al. published this in Nature in 2024 under a title that doesn’t bury the lede: AI models collapse when trained on recursively generated data.[^2] Not the popular middle — the edges. Niche dialects. Minority opinions. The one historian who disagreed with the consensus. The post that nobody linked to but happened to be right.

Each generation eats its own tail. The long tail goes first.

Call this recursive forgetting. It’s not malicious. It’s not even a bug. It’s the geometry of training on a distribution that no longer matches reality. And the share of synthetic training input is forecast to dominate by the end of the decade — Gartner has projected synthetic data will overshadow real data in AI training by 2030.[^3]

After enough cycles, the model’s worldview converges to its own prior. What was outside that prior is no longer reachable, no matter how true it was.

one percent of classical literature survived the last index

We’ve seen this before, on a slower timescale.

Roughly 1% of classical Greek and Latin literature survived to today.[^4] Of about three hundred Athenian tragedies known by name, thirty-two complete texts came down to us. The cause, contrary to the school-textbook narrative, was almost never a single fire. The Library of Alexandria didn’t fall in one night. It was indexing decisions, copying priorities, taste shifts, storage failures, compounding over centuries. The texts that didn’t get re-copied didn’t survive. The texts that didn’t get translated didn’t get re-copied.

Survivorship was determined upstream of any reader. It was a function of which scribes thought a text was worth their week.

The same machinery is running again. Different scribes.

most books ever written are not in any model

Before we even get to the indexing question, there’s a more boring one: a lot of human writing simply isn’t digitized.

Google estimated in 2010 that there were about 130 million distinct book titles in the world.[^5] Fifteen years later, Google Books has scanned roughly 40 million; HathiTrust, the largest open scholarly mirror, holds about 18 million volumes.[^6] Counting the overlap, well under a third of everything ever published sits in any digital corpus an LLM can train on.

The rest sit in physical libraries, university basements, private collections. For a model, they aren’t just hard to access — they don’t exist. The author’s argument was made, the book was published, and the model has never heard of it.

Most of human knowledge, from the model’s perspective, is already gone.

the archive is one lawsuit thick

The thin part of the corpus is also fragile.

In June 2024, the Internet Archive removed about 500,000 books from its Open Library after losing Hachette v. Internet Archive at the district court level.[^7] In September 2024, the Second Circuit affirmed.[^8] One of the largest non-commercial digital archives in the world lost half a million titles to a single ruling.

The Internet Archive is one organization. The Wayback Machine — the only meaningful record of what the open web used to look like — depends on the same nonprofit budget, the same legal exposure, the same single point of failure. There is no second archive of the same scale.

If the next ruling goes the same way, or the funding cliff hits, or the building floods, that record goes with it. The model doesn’t need to be told the books are gone. They just stop being in the next training run.

censorship without a censor

Pull these threads together and a quieter pattern shows up.

You don’t have to ban a book. You only have to keep it out of a few key sources before the next training cycle. You don’t have to silence a writer. You only have to keep their site out of the index. You don’t have to rewrite history. You only have to wait for the long tail to thin, and let recursion do the rest.

This is censorship in the mathematical sense — a reduction in the support of the distribution — without anything that looks like a censor. No fire, no banned-books list, no public villain. Just an indexing decision, propagated.

And because all of this happens upstream of the reader, the reader can’t tell. You don’t notice the books that aren’t on the shelf when you’ve never seen the shelf.

four models read it for everyone

You don’t need most of the world’s content to be missing. You only need it to be missing from three or four models.

People use what’s good and convenient. The LMArena leaderboard is dominated by a handful — GPT, Claude, Gemini, Grok — and the next thirty are footnotes for hobbyists. If you ask a real person a real question this year, the answer almost certainly came through one of four models.

Concentration is the multiplier. If your work isn’t in the corpus those four were trained on, or it sits below their relevance threshold, the median reader of 2030 will never encounter it — even if the page still exists, even if Common Crawl still has a snapshot somewhere. The funnel is narrow on purpose. Better models cost more to train and serve. The economics push toward a few winners. The few winners pick what survives.

what this means for your work

If you write on the internet today, you’re competing for a slot in a small number of models’ worldviews. Most of what gets published doesn’t get one.

The Substack post you wrote last March. The Notion site about your startup. The LinkedIn essay that did 200 likes. Your work probably won’t vanish — most of it gets scraped eventually. What happens is harder to fight: it gets diluted into a sea of AI-drafted content written around the same prompts. The model knows you wrote something. It just doesn’t weight you above the noise.

And if you don’t make the next training cycle at all, or you make it but get down-weighted to background, the practical effect is the same as if you’d never published. The reader asking the model never finds out you wrote it.

You can’t control the index. You can control what you put in it, and how often. You’re either in the index, or you’re already invisible.

a record, in case

The probability that this very post survives the next decade in any LLM’s worldview isn’t zero. But it isn’t one either.

Neither is yours.

This blog is a small bet against erasure — a place to leave what I think and what people I admire think, in stable URLs, mirrored, fed to the index, so that some version of it has a chance to land in whatever the readers of 2035 are using.

I don’t think the internet has to go dead. The library can be rebuilt — better this time, because we know what’s at stake. This is a challenge to solve, not a fate to accept.

The default outcome is erasure. The exception is what you choose to put in the index, and how often. Write the thing. Publish it somewhere stable. Mirror it.

[^1]: Graphite, More Articles Are Now Created by AI Than Humans, Oct 2025 — n = 65,000 English-language Common Crawl articles, Jan 2020 – May 2025, scored with Surfer’s AI detector. The 14% Google-Search visibility figure is from the same study. [^2]: Shumailov, I., Shumaylov, Z., Zhao, Y., Papernot, N., Anderson, R. & Gal, Y. AI models collapse when trained on recursively generated data. Nature 631, 755–759 (2024). [^3]: Gartner has projected for several years that synthetic data will overshadow real data in AI training by 2030. Earlier waypoint: Gartner (2021) projected 60% of AI/analytics data would be synthetically generated by 2024. [^4]: Standard scholarly estimate — see Reynolds & Wilson, Scribes and Scholars — though contested as to method. Of ~300 Athenian tragedies known by name, 32 complete texts survive. [^5]: Leonid Taycher, “Books of the world, stand up and be counted! All 129,864,880 of you,” Google Books blog, August 2010. [^6]: Google Books reported >40M scanned at the 15-year anniversary in October 2019. HathiTrust reported >18M volumes in September 2024. [^7]: Internet Archive blog, “Let Readers Read,” June 17, 2024 — cites the ~500,000 figure for books removed from CDL after the publishers’ victory at the SDNY. [^8]: Hachette Book Group, Inc. v. Internet Archive, No. 23-1260, 2d Cir. Sept. 4, 2024 (affirming SDNY).

                                    .    ..   .........       ...
                                  .. ..:::==:.          ....:=-::-.
                                    :-:....=%%*+=-:::::::+%*#= .=-..
                                ...=.       :%@@%-::.....##+. =+:
                               .  :%+:...  . .#@#.    . -@+. ==. .
                              . .-:-+*-...  ..:@%:......#*. -= ..
                             ..-+.   .::..  .. =@: ... -%. := ..
                             . =*-: ........  . #+ ... *= .-...
                                .-+-. ... .:-.. -#.... -. -. .
                               .  .=*+.  .. ==.. =: ...  :- .
                                 .  :#%=:  .:*.=: .. .- .+. .
                                   . .#@%=. -@+.=-...#+.==:..
                                    . :###**#@%::---=*=----..
                                    ...+-----====---:-:-=-: .
                                    ..:+-==-++=---:=-==+-+= ..
                                   .. =##+-+@+.....:::=@+--:  ..
                                  .  -+%+  **        . :*=.:-:  .
                                 . :+=-=  -+ .......... .--..--.  .
                                . -*=....-= .......    .. ... .=-. .
                              .. =+:  ...- ........:-: ...  ..  :+: .
                              ..==.. .. .. . ..   .%@+.   ..  .. +@- .
                             . ==.. ..   .     .-*#@@@%%+:     ..:*@- .
                            . -= .  ..      . :*@%+%@*=*@@*..   . .*%...
                           ...+... ..       ..+@#. #@:  .-:..   .. :%+ .
                           . +: ..... ..    . +@+  #@+..        ... -%:..
                          . :+ .  ..  ..    ...#@+.*@+ ....     ....:** .
                          . =- . ..   ..     . .*@%%@*.  ..      ....+%. .
                           .+. . ..  ....   ...  :-#@@%*-  .    .... :%= .
                         . := .  .. ...... ..  ..  -@#+@@*. .   ... ..+# .
                         . =- .     .....  ..    ..:@#.:*@#: .   .. ..=*.
                         . +: .  .   ...   ..  ... .#%- .%@+ .   .. ..-+
                          .+:.. ..    ..  ...     ..#@+. =@#. .     ..:+
                          .+-.. ..       ......:=.  +@*  *@#..   .....-= .
                          .+-:......      .....*@@#**@#=*@#- .  .. ..-*:..
                         ..=........   .. ...   :=+*%@@%+=. .  .. ...**.  .
                        . :#...  ..  . -. ..  ..    :%@=   . ... :: :*#*+: .
                       . .*+:: . ..  . =- ..   ......::.........:-... ..-*=..
                      . :*=-. . .    . -+ .     ..      ...  ...: :=.    .=- .
                     ...#=.. .+=.... ..:*= ..   ... ....   .:.   -%- ..... +.
                      ..#-. .-+%#=:::. :*%-  ...  ....  .-**-:=*@@#-:---:---..
                      . :*=-+##%@%+--:-+#%@+:...    ..:+#%%*+==+==---:::::.
                       . .---=-:.....::--==+=--:::::::--:..                .
                        .                     .......      ................
                         .........  .........        ......

Most founders pick investors by brand. We picked ours by Venn diagram.

When Ezequiel Sculli (my Darwin co-founder) and I went out to raise our pre-seed, we sat down and listed two things on a whiteboard: what we were already great at, and what we knew we lacked. The first list was long enough to feel comfortable. The second list told us exactly which checks to chase.

what we had

We’re builders. Capitalize us correctly and we ship: we find problems people pay for, we put together teams, motivate them, delegate, manage the boring middle. That’s been true at Sirena, it’s true at Darwin. Building is the part of the job we know.

what we didn’t have

Two gaps, named honestly:

1. Fundraising muscle. Neither of us is a born fundraiser. We close founders, customers, and operators all day. We don’t naturally enjoy running a 60-investor process, and we needed people on the cap table whose job it is to be aggressive about the next round.

2. Country-specific connections. We knew Latam mid-market because we’d shipped to it before. We didn’t have enterprise networks in every country we wanted to operate in, and we had thin US presence — both for late-stage capital and for the AI-frontier conversation in San Francisco.

Two gaps. So we drew three circles.

the three circles

• Building. What Ezequiel and I already had.

• Fundraising. The muscle to import.

• Connections in Latam + US. Geographic reach, country by country.

Three circles. We sit in the first one. Every investor we picked had to land somewhere in the other two.

the fills, round by round

Pre-seed (Feb 2024, $2.5M, Canary led):[^1]

• Canary (Brazil) — fundraising muscle plus enterprise connection in our biggest market.

• Bridge — now Nazca after the 2025 merger[^2] (Mexico) — local presence in MX.

• Dalus Capital (Mexico) — second MX connection layer.

• H20 Capital Innovation (Mexico + Miami) — bridge between MX corporate and US LP capital.

• Latitud (Latam → US) — the funnel from Latam operators to US capital. They’re not just money; they’re a network of LP relationships.

• FJ Labs (NYC) — Fabrice Grinda’s fund. NYC presence, marketplace pattern recognition, and a fundraising muscle that has compounded over decades.

Seed (Aug 2025, $4.5M, Base10 led):[^3]

• Base10 Partners (San Francisco) — the SF + AI-frontier connection layer. Base10 sits inside the conversation that decides which AI companies are real and which are demos. We needed that reach to keep building inside the wave, not adjacent to it.

Each name above is in our cap table because they fill a specific region of the Venn. None is there for the brand alone.

the methodology

If you’re a founder about to raise, run the same exercise:

1. List what you and your co-founder are individually great at. Be honest. Strengths only — no aspirational lines.

2. List what’s missing. Geography, function, network, fundraising aggressiveness, public profile, regulatory access — whatever you actually need but don’t have.

3. Name each gap as a circle.

4. Pick each investor to fill a circle, or an intersection.

Don’t pick investors by brand. Pick them to complete the team you and your co-founder aren’t.

The cap table is the most permanent team you’ll ever build. Build it like a team.

[^1]: TechCrunch, Darwin AI gives small LatAm companies AI-powered sales assistant, Feb 26, 2024 — confirms the $2.5M pre-seed and the lead-investor list. [^2]: Bloomberg / LatamList, Latam VCs Nazca and Bridge Merge to Combine $300 Million in Assets, March 2025. [^3]: Axios Pro / FinSMEs / LatamList, Darwin AI raises $4.5M seed round led by Base10 Partners, Aug 2025.