LatAm Tech Weekly

#244: AI and Marathons, Kimi K3, deals of the week... and much more!

Weekly writing about what is happening in LatAm tech. By day, I am part of the corporate development team at Itau Unibanco. By night, I am reading and learning about technology in general (now, with a focus on AI). During the weekends, I’m writing the LatAm Tech Weekly. And obviously, always running!

If you have not subscribed yet, join the 14,700+ weekly readers by subscribing here!

Happy Sunday!

First things first: I’m off on long-awaited vacation!! Ten days off — or, well, “off.” If you’re an avid reader, you already know how this goes: I swear I’ll skip an edition, and then inspiration strikes me mid-vacation and I end up writing anyway. So — let’s see what this trip has in store... But without further ado, let’s dig in!

Follow me on LinkedIn , Instagram or X for daily updates!

Opinions expressed here are solely my own and do not represent those of people, institutions, or organizations that I may or may not be associated with in any capacity, unless explicitly stated.

AI Won’t Replace the Marathoners

There is a moment in every marathon — usually somewhere around kilometer 30 — when the race stops being about talent or training. Your legs are heavy, the glamour is gone, and whatever carried you through the first 30km simply stops working. What gets you to the finish line is not a breakthrough — it is everything you did in the boring, unglamorous long runs that nobody saw.

I’ve been thinking about that moment a lot lately — because I think AI just hit its kilometer 30 (and obviously because I am training for my third marathon this year - below a picture of my latest 24km long run!)

I read two pieces this week that when read together, make the case better than either does alone. The first is David Brooks’ essay in The Atlantic, “The People Who Will Thrive in the AI Age” (June 2026). The second is the Work AI Index 2026 — a survey of 6,000 knowledge workers by Glean’s Work AI Institute, dissected at length on The AI Daily Brief (the podcast many of you know I listen to daily). The pieces are very different, but they arrive at the same place.

Volition is the new scarcity

Brooks’ core argument fits in seven words: when intelligence is plentiful, volition is valuable.

But before he gets to volition, he draws a map — and the map deserves a pause, because the entire essay rests on it. What separates people in an AI-saturated world, Brooks argues, is not intelligence. It is their relationship to mental effort. And on that spectrum, he sees three archetypes.

At one end sit the **cognitive misers** — the people who find thinking hard genuinely unpleasant and take any opportunity not to do it. For them, AI is less a tool than a permission slip: a machine that thinks so they no longer have to. They outsource judgment happily, accept the first output, and move on with their lives.

At the other end are the people psychologists describe as having a **high need for cognition** — the ones energized by hard problems, who play difficult games and read dense books for fun, who treat AI as a sparring partner rather than a crutch. Brooks gives them a name that, I confess, I personally related: **mental marathoners**. People who insist on running the mental marathon the way some of us insist on running 42km even though the car exists. His bet is that this group will actively resist handing over the thinking — not out of technophobia, but because they want their work to remain original, personal, theirs. They use AI to expand their agency, not replace it.

And in the middle sits everyone else — the **medium need for cognition** majority, who could tip either way. To me, this is the most consequential group in the essay. Brooks makes the point that need for cognition is not a fixed trait; willpower is extremely sensitive to context. Which way the middle drifts will decide whether AI produces what he calls extreme cognitive polarization — a world split between minds that atrophy and minds that compound.

As an actual (not just mental) marathoner, I felt personally summoned… Brooks’ warning is sharp. In one study he cites, an AI intentionally gave incorrect answers — and humans accepted them as true 80% of the time. He calls this failure mode cognitive surrender. But what makes his argument especially relevant here is the parallel he draws to endurance. Thriving in the AI age, he argues, will look less like a burst of brilliance and more like sustained intellectual effort — the willingness to keep engaging, questioning, and refining when it would be easier not to. In his framing, the winners are not those who access intelligence fastest, but those who stay in the work longer. The ones who don’t hand over the thinking at kilometer 30.

In other words: when everyone has access to the same intelligence, what differentiates people is no longer how smart they are. It’s how willing they are to keep doing the work.

The productivity paradox, quantified

Now let’s look at the data. The Work AI Index 2026 surveyed 6,000 knowledge workers between December 2025 and January 2026, and the headline numbers are the kind you’d put on a keynote slide: 87% of knowledge workers now use AI. 73% say it makes them more productive. Reported time savings run around 11 to 13 hours per week — nearly a third of a standard workweek.

And then the shock: only 13% strongly agree their organization is performing significantly better because of it. Billions of hours saved — and almost none of it visible at the organizational level. Why?

Part of the answer is a phrase we’ll likely hear a lot more: bot sitting. The report defines it as the hidden human labor required to make AI usable — feeding context, checking outputs, debugging mistakes, re-running prompts, and cleaning up confident-but-wrong answers. Workers spend an average of 6.4 hours per week on it. Roughly half the productivity dividend gets reinvested into supervising the tool that produced it. The report also found that 36% of AI sessions fail outright, forcing a restart or heavy rework.

It gets darker. Fatigued workers eventually lower their standards — the report calls the result bot slop — and around 40% of employees admit to shipping AI-generated work they could not explain or defend if asked. And when AI-generated work fails, heavy users are 3.4x more likely than light users to blame the tool rather than themselves.

Cognitive surrender, measured.

The AI Daily Brief framed this in a way that stuck with me: we keep asking “how can AI help me do the same work faster?” when the better question is “what can I now do that was previously impossible?” Individual time savings do not automatically compound into organizational transformation. That requires redesigning the work itself — which is training, not shopping. Buying a better watch has never made anyone run a faster marathon…

Meanwhile, the front of the pack accelerates

The paradox becomes sharper against the backdrop of what’s happening at the frontier — because even by recent standards, the past week was absurd. What’s driving it is the same shift underneath everything else in this piece: the unit of consumption is no longer a seat. It’s a token. A developer running an agent against a real codebase can consume hundreds or thousands of dollars a month — a number no per-seat pricing model ever contemplated.

Kilometer 30, enterprise edition

Which brings us to my favorite data point of the quarter: Uber blew through its entire 2026 AI budget in four months. Read that not as retreat but as maturation. Budgets, dashboards, governance, exception processes — this is what happens when a technology stops being an experiment and becomes a line item. The sprint phase is over. The conversation inside enterprises has moved from access to allocation, from novelty to ROI.

And here is the uncomfortable part: Uber’s COO admitted it remains genuinely hard to draw a line from all that AI usage to more product actually shipped. The Glean data explains why. Adoption is solved. Transformation is not.

We may be entering a new phase where the first enterprise constraint is not access to AI, but unbounded consumption. Tokens are cheap individually but expensive in aggregate — and without clear task redesign, they scale activity faster than they scale outcomes. Thousands of prompts to refine a slide, hundreds to draft a single email, fleets of loosely coordinated agents producing work that still requires human verification.

The emerging question is not just how many agents can we deploy, but how much work should exist in the first place — and how to harness, orchestrate, and constrain that work so that it compounds into real output rather than noise. In other words: the problem is no longer intelligence scarcity. It is discipline.

The people who keep running

So what actually separates the 13% of organizations where AI gains show up from everyone else? The Work AI Index has a beautiful answer buried in it. High performers don’t bot-sit less — they bot-sit differently. They treat the friction as training: they are more than twice as likely to describe AI as a valuable teacher, and they reinvest saved hours into building new skills rather than simply producing more output. Adoption spreads through example, not mandate — seeing a cross-functional colleague use AI well makes the average employee 5x more likely to adopt, precisely because those workflows have survived contact with real organizational messiness.

That is not a technology story. It is a training story.

Long-distance running taught me that performance is never the product of one heroic effort — it is thousands of unglamorous marginal gains, compounding quietly on days when motivation is nowhere to be found. The models will keep getting better, cheaper, and more agentic whether we show up or not. The frontier labs are running their own race, at a pace none of us can influence. Ours is different.

Nearly everyone now has access to frontier intelligence — which means intelligence is no longer the differentiator. Volition is. Judgment is. The willingness to redesign how you work, verify what the machine gives you, and keep learning after the novelty wears off — that is the race that is actually winnable…

The best marathoners are not the ones who find shortcuts. They’re the ones who keep running!

General news:

• Anthropic extended free access to Claude Fable 5 until July 19 before introducing usage-based pricing for paid subscribers. The move reflects the rising compute costs of frontier AI models and signals a shift toward API-style monetization for premium AI services. 🇺🇸

• More than 200 economists and technology leaders published a manifesto warning that AI could reshape the global economy more profoundly than the Industrial Revolution. The group calls for new institutions and public policies to ensure AI complements human labor while mitigating the risk of large-scale workforce disruption. 🌎

• Brazilian digital lending fintechs originated a record R$53.8 billion in credit in 2025, a 51% increase year over year. The data indicates the sector is entering a more mature phase, with growth increasingly driven by portfolio expansion rather than new products. 🇧🇷

• Lions Startups plans to incubate 100 startups over the next five years through its venture builder model in Paraná. The initiative aims to strengthen entrepreneurship outside Brazil’s traditional innovation hubs by providing infrastructure, mentorship and investor access in exchange for equity. 🇧🇷

• Venture capital fundraising in the U.S. became increasingly concentrated, with Andreessen Horowitz, Thrive Capital and Founders Fund raising a combined US$34.8 billion in the first half of 2026. The figures highlight investor preference for large AI-focused managers while fundraising for first-time funds fell to its lowest level in a decade. 🇺🇸

General news:

• Meta is facing a lawsuit from 26 former employees who allege AI-driven performance evaluation tools disproportionately targeted workers with disabilities or on medical leave during layoffs. The case could become a landmark legal test for the use of AI in workforce management. 🇺🇸

• DeepSeek is preparing a new funding round and a potential IPO at a reported US$71 billion pre-money valuation. The AI startup plans to invest heavily in proprietary data centers and computing infrastructure as competition among frontier AI companies intensifies. 🇨🇳

• Brazilian cleantech Lemon Energia is expanding into the residential market while evaluating acquisitions to consolidate the sector. The company aims to capitalize on Brazil’s electricity market liberalization and reach millions of new consumers. 🇧🇷

• Vakinha is strengthening its crowdfunding platform after adapting to the rise of Pix with enhanced fraud prevention, faster payouts and new fundraising tools. Serving more than 15 million users, the company continues to expand its ecosystem while reinforcing trust as its primary competitive advantage. 🇧🇷

Deals:

• IM8 secured a US$1 billion customer financing facility from General Catalyst’s Customer Value Fund. The non-dilutive structure finances customer acquisition costs instead of taking equity, highlighting the growing adoption of revenue-linked financing for subscription businesses. 🇬🇧

• Uruguayan mobility startup TripWip raised a US$4.2 million seed round to expand across Mexico and Argentina. The funding will support AI development, marketing and geographic growth as the company scales its peer-to-peer car-sharing platform. 🇺🇾

General news:

• Pix has become a central issue in the U.S.–Brazil trade dispute, with the Trump administration arguing that Brazil’s instant payment system disadvantages U.S. payment companies such as Visa and Mastercard. Brazil rejects the claim, maintaining that Pix is a public financial infrastructure designed to foster competition and financial inclusion. 🇧🇷🇺🇸

• Senior Sistemas is accelerating its M&A strategy after acquiring ERP provider Cigam and HR tech Salú for a combined R$481 million. The company expects acquisitions to contribute R$2 billion toward its R$5 billion revenue target by 2032 while expanding AI capabilities across its enterprise software platform. 🇧🇷

Deals:

• Neko Health, the preventive healthcare startup co-founded by Spotify founder Daniel Ek, raised a US$700 million Series C led by Lightspeed Venture Partners to expand into the U.S. and scale its AI-powered preventive healthcare clinics. The company now serves 100,000 members and has raised nearly US$1 billion to date. 🇸🇪🇺🇸

• Stripe and Advent International submitted a US$53+ billion bid to acquire PayPal, offering a 28% premium over its pre-offer share price. If completed, the transaction would rank among the largest fintech acquisitions ever and significantly reshape the global digital payments industry. 🇺🇸

• Peruvian foodtech +NUTRI Co. was acquired by Uruguay’s Terraflos in one of Peru’s largest startup exits in recent years. The deal will support regional expansion and the launch of new AI-designed functional nutrition products across Latin America. 🇵🇪🇺🇾

• Mexican construction fintech Mango extended its seed round to US$6 million, led by Brick & Mortar Ventures, to expand its financial infrastructure platform for the construction industry. The company will launch new financial products and accelerate expansion across Mexico and Latin America. 🇲🇽

• Quick Soft acquired ERP provider RGBtec, its second acquisition in less than 60 days, strengthening its position in Brazil’s receivables management market ahead of the rollout of the electronic duplicate (duplicata escritural). The company aims to surpass R$1 trillion in annual transaction volume by 2030. 🇧🇷

General news:

• Visa launched the Visa Stablecoin Platform (VSP), enabling banks, fintechs and payment providers to issue, custody and transfer stablecoins through Visa’s infrastructure. The platform integrates with OpenUSD, whose consortium includes Itaú, Bradesco, Mastercard, Stripe and BlackRock, strengthening institutional adoption of blockchain-based payments. 🇺🇸🌎

• B3’s Regime Fácil has facilitated four corporate debt issuances totaling R$149 million in its first four months, helping smaller Brazilian companies access capital markets through a simplified issuance framework. The initiative is designed to lower costs and expand financing alternatives for mid-sized businesses. 🇧🇷

• Yungas launched a R$5 million investment program for AI startups serving franchise operations, making its first investment in Fin, an AI-powered financial management platform. The initiative aims to automate financial workflows across franchise networks and expand AI adoption in retail operations. 🇧🇷

• Primus Ventures began investing through its new R$100 million Sul Ventures fund, targeting approximately 20 B2B startups in Southern Brazil with initial investments ranging from R$1 million to R$5 million. The fund focuses on pre-seed and seed-stage companies while expanding Primus’ regional venture capital strategy. 🇧🇷

Deals:

• Uber agreed to acquire Delivery Hero for US$14.8 billion, creating one of the world’s largest food delivery platforms across 99 countries. The transaction also includes the exit of Prosus from Delivery Hero and selected European asset divestitures to address antitrust concerns, marking another major consolidation move in global food delivery. 🇺🇸🇩🇪

• Fireworks AI raised US$1.5 billion at a US$17.5 billion valuation after surpassing US$1 billion in annualized revenue. Backed by Nvidia, the AI infrastructure company will expand GPU capacity and enterprise operations as demand for open-source AI models accelerates. 🇺🇸

• Canopy entered Brazil’s agribusiness software market through the acquisition of Maxicon Sistemas, whose software supports logistics and grain operations representing roughly 10% of Brazil’s soybean production. The acquisition strengthens Canopy’s vertical software consolidation strategy following its US$100 million funding round. 🇧🇷

General news:

• Apple became the world’s most valuable public company again, reaching a market capitalization of approximately US$4.88 trillion after Nvidia shares fell amid a broader AI semiconductor selloff. Investors are increasingly favoring companies expected to monetize AI through ecosystems and software rather than solely AI infrastructure. 🇺🇸

• Moonshot AI unveiled Kimi K3, a 2.8 trillion-parameter open-source model that reportedly rivals leading U.S. AI models at a fraction of the cost. The announcement renewed concerns over AI infrastructure spending and triggered a selloff in semiconductor stocks. 🇨🇳

• Agentic payments are moving closer to commercial adoption, with Visa and Mastercard enabling AI agents to search, negotiate and complete purchases on behalf of users. Brazil’s Central Bank is expected to evaluate a regulatory framework as companies explore integrating autonomous payments with Pix. 🇧🇷🌎

• Brazilian payments company Elo is expanding beyond cards with biometric Pix, introducing biometric authentication for instant payments while positioning itself for future AI-powered commerce and digital identity services. 🇧🇷

• A Galileo report ranked Brazil as Latin America’s leader in personalized digital banking, with half of financial institutions reaching advanced personalization capabilities, supported by Open Finance, embedded finance and real-time payments infrastructure. 🇧🇷

• Chilean fintech Toku is accelerating its expansion in Brazil, targeting a sixfold increase in local revenue by leveraging Open Finance, Pix Automático and AI-powered recurring payment automation. 🇨🇱🇧🇷

Deals:

• Databricks reportedly raised a new funding round at a US$188 billion valuation, reinforcing its position as one of the world’s largest AI infrastructure companies. The capital will support AI product development, acquisitions and expansion ahead of a potential IPO. 🇺🇸

• Anthropic is in talks to lease up to US$10 billion of AI computing capacity from Meta over two years, allowing Anthropic to diversify compute providers while helping Meta monetize its growing AI infrastructure investments. 🇺🇸

• Argentine fintech Ualá secured an additional US$20 million investment from Tether, bringing the stablecoin issuer into its shareholder base as part of its US$197 million funding round. The capital will support expansion across Argentina, Mexico and Colombia. 🇦🇷🇲🇽🇨🇴

• Banco Inter completed a R$300 million subordinated debt issuance, strengthening its regulatory capital base through Tier 2 instruments while supporting future lending growth. 🇧🇷

The biggest news in AI this week comes out of China: Moonshot AI released Kimi K3, a 2.8 trillion-parameter model that is now, by a wide margin, the largest open-weight model ever shipped. It matters because it isn’t just big — it’s genuinely competitive, using a mixture-of-experts design that splits itself into 896 smaller expert subnetworks and activates only a portion for any given task, letting it hit near-frontier performance without frontier-level compute costs. On the numbers that matter: a 1 million token context window, with Kimi’s “Delta Attention” enabling up to 6.3x faster decoding at that scale, and — this is the part that stings — early head-to-head benchmarks have it beating Anthropic’s Fable 5 in 7 of 8 coding arenas, with Fable’s only real edge being speed. Pricing seals the disruption story: K3 lists at $3/$15 per million tokens (input/output) versus Fable 5’s $10/$50 — a fraction of the cost for benchmark parity. Full weights don’t drop until July 27th, so for now it’s API/subscription-only via kimi.com, but the timing — landing right before the World AI Conference in Shanghai, weeks after the US briefly pulled Fable and Mythos over export-control concerns — reads as a deliberate flex: proof that the open-vs-closed gap has narrowed to about one generation, not several.

• Febraban Tech 2026
  Date: August 24–26, 2026
  Location: São Paulo, Brazil
  Description: One of the main financial technology and innovation events for the banking and financial services sector in Latin America.
  More info

• Conta Azul Con 2026
  Date: August 1, 2026
  Location: São Paulo, Brazil
  Description: A conference connecting accounting, entrepreneurship, and technology through keynotes, workshops, networking sessions, and discussions on business innovation.
  More info

• Rio Innovation Week 2026
  Date: August 4–7, 2026
  Location: Rio de Janeiro, Brazil
  Description: One of Brazil’s largest innovation festivals, bringing together entrepreneurs, startups, corporations, investors, and policymakers to discuss technology, business, and social transformation.
  More info

• SP2B (São Paulo Beyond Business) 2026
  Date: August 9–16, 2026
  Location: São Paulo, Brazil
  Description: A new business and innovation event designed to connect entrepreneurs, executives, investors, and corporations through discussions on growth, creativity, and the future of business.
  More info

• Deep Tech Summit 2026
  Date: August 11–12, 2026
  Location: São Paulo, Brazil
  Description: A conference focused on science-based innovation, bringing together deep tech startups, investors, corporations, and researchers to explore frontier technologies and emerging opportunities.
  More info

• Startup Summit 2026
  Date: August 26–28, 2026
  Location: Florianópolis, Brazil
  Description: One of Latin America’s leading startup events, connecting founders, investors, and ecosystem leaders through content, networking, and business opportunities.
  More info

• HackTown 2026
  Date: September 3–7, 2026
  Location: Santa Rita do Sapucaí, Brazil
  Description: A unique innovation, technology, and culture festival that transforms an entire city into a hub for networking, learning, entrepreneurship, and creative collaboration.
  More info

• Hotmart Fire 2026
  Date: September 10–12, 2026
  Location: Belo Horizonte, Brazil
  Description: One of the leading events for the digital business and creator economy ecosystem, covering marketing, sales, online education, innovation, and business growth.
  More info

• Dreamforce 2026
  Date: September 15–17, 2026
  Location: San Francisco, United States
  Description: Salesforce’s flagship conference focused on CRM, AI, digital transformation, customer experience, and enterprise innovation.
  More info

• TechCrunch Disrupt 2026
  Date: October 13–15, 2026
  Location: San Francisco, United States
  Description: A leading global startup conference where founders, investors, and technology leaders discuss entrepreneurship, fundraising, and innovation.
  More info

• Web Summit Lisbon 2026
  Date: November 9–12, 2026
  Location: Lisbon, Portugal
  Description: One of the world’s largest technology conferences, bringing together entrepreneurs, investors, executives, and policymakers to discuss global innovation trends.
  More info

• Slush 2026
  Date: November 18–19, 2026
  Location: Helsinki, Finland
  Description: A globally recognized startup and venture capital conference focused on connecting founders and investors while fostering innovation and growth.
  More info

• AWS re:Invent 2026
  Date: November 30 – December 4, 2026
  Location: Las Vegas, United States
  Description: AWS’s flagship cloud computing conference, featuring product launches, technical sessions, training, and discussions on cloud infrastructure, AI, and data.
  More info

• The Midnight Feast - Lucy Foley

• (the articles mentioned above!)

• The AI Daily Brief

“The miracle isn’t that I finished. The miracle is that I had the courage to start.” — John Bingham