Mind the gap
The distance between what AI can do and what it's actually doing is the only number that matters.
Hey there,
I’m just coming back from my golf practice. I spent most of the winter working on my golf swing.
I’m about a 2 handicap. On a good day, I can shoot par, which is better than most amateurs but still far from pro level. Last year was my best year yet. Even so, I came back in January and started over from scratch.
I focused on where my hands are during the takeaway. It sounds minor, but getting it right means relearning the whole movement. There’s a comfortable way to do it, which is an easy improvement but only brings a small gain. The harder way feels wrong for months, but eventually leads to something bigger.
I keep thinking about that at work. Lupa is at about $2M in revenue. To get to $20M, I need to change focus completely: how I lead, what I hand off, what I stop doing myself. None of it is comfortable. All of it is necessary.
When hiring, I keep asking clients: Are these roles right for where you’re going, or just leftovers from when you were smaller? Most founders inherit job descriptions from earlier versions of their company. It’s easier to keep filling those roles than to stop and ask if they still fit.
There was a lot of news this week, but only a few stories actually mattered.
Let’s get into it.
🌐 News Shortlist
1. Anthropic publishes real data on which jobs AI is displacing; the most important finding is buried beneath the headlines.
Recap: On March 5, Anthropic released a labor market paper with a new metric: observed exposure. It tracks what AI could do versus what people actually use it for at work. The most exposed jobs by real usage data are computer programmers, customer service reps, and data entry keyers. Financial analysts and market research specialists follow. The least exposed—cooks, lifeguards, dishwashers, bartenders—make up about 30% of the workforce. Those in the most exposed roles are more likely to be female, older, more educated, and higher paid. Despite fears, there hasn’t been a systematic rise in unemployment for high-exposure jobs since ChatGPT launched. But there’s a telling signal: entry-level hiring for younger workers in exposed fields has dropped 14% since late 2022.
Commentary on the report is missing the real gap: not between safe and exposed jobs, but between what AI can do and what it’s actually doing. For example, computer and math jobs have 94% theoretical exposure, but only 33% actual coverage. Office and admin jobs are 90% theoretical, 25% actual. Legal is 80% theoretical, 15% actual.
That gap isn’t a comfort. It’s a countdown, showing where displacement is heading and how much runway is left. Anthropic’s framing is cautious: effects so far are modest, but the mechanism is now in place and the trajectory is clear.
The drop in entry-level hiring for younger workers is the clearest signal. There is no mass unemployment yet, but a 14% drop is how disruption begins: quietly, at the edges, before it appears in the data. Companies aren’t firing people. They’re just not replacing them or hiring new graduates to fill the pipeline.
This explains why the international hiring market looks different right now. In Latin America, demand is up for roles needing high judgment and direct client contact—the kinds of things AI can’t handle yet. Task-based roles are slowing, matching Anthropic’s data almost perfectly.
What matters here is that Anthropic, the company behind Claude, isn’t a neutral observer. Publishing data that implicates their own product in labor market disruption is meaningful, and deserves credit.
Advice:
Read the report itself. The chart showing theoretical versus observed exposure by job category is the best way to see where things are going. If much of your team’s work is in the “blue zone,” meaning what AI could do but isn’t doing yet, that’s your planning horizon.
2. Latin America’s startup funding just had its strongest year since 2022, but fewer companies benefited.
Recap: The Latin America VC Report 2026 found venture capital investment hit $4.1 billion in 2025—a 13.8% rise over 2024, marking the first real recovery after three years of decline from the 2021 peak. Still, the number of deals dropped to 681, the lowest since 2017. Average check size grew to $6.1 million. Brazil and Mexico took most of the capital. Fintech dominated funding, but late-stage rounds shrank sharply, and seed deals fell from 321 to 247.
The important number isn’t the $4.1 billion; it’s how few deals actually happened.
More capital is going to fewer companies, raising the bar for getting funded. Investors write bigger checks only for those who have already executed. The speculative bets of 2021 are gone. Now, a smaller set of investors are concentrating capital on proven founders in established markets. Brazil and Mexico took most of it. Fintech still led. Seed-stage startups across the rest of the region are struggling.
This shift changes how startups build teams. When money is easy, founders hire to show momentum. Now, with capital concentrated, they hire to prove execution. These are different philosophies, with different roles and candidates. The companies still getting funded can show every hire made a difference.
The real warning sign is the drought in seed-stage funding. Seed deals dropped sharply in a year. That’s the pipeline for future Series A rounds—if the seed market stays tight, many startups simply won’t exist to raise later. A thinner early ecosystem will ripple through hiring, talent, and the health of LatAm’s tech workforce before it shows up in big-picture VC data.
There’s one reason for optimism: several VC firms that raised funds in recent years are close to deploying that capital, so a new wave of investment is likely in 2026 and 2027. The rebound has started. The only question is whether it broadens out or stays concentrated at the top.
Advice:
If you’re a founder raising in Latin America, every hire needs a clear answer to what they make possible that no one else can. Investors are grilling teams the same way they question products. Build for that scrutiny.
3. Oracle plans to cut up to 30,000 jobs to fund its AI data centers.
Recap: Bloomberg reported on March 5 that Oracle is planning major layoffs across divisions, with estimates as high as 30,000 jobs, which is about 12 to 18 percent of its global workforce. The reason is a cash crunch from a massive AI data center expansion, including a $156 billion OpenAI deal. Oracle’s stock has dropped sharply, open positions are frozen, and they’re considering selling the Cerner healthcare unit. Wall Street expects negative cash flow for years before the AI investment pays off.
This is a different kind of layoff than what happened at Block.
Block cut people because AI made them more productive. Oracle, in contrast, is cutting people to pay for an overextended bet on AI infrastructure. These layoffs aren’t about efficiency; they’re collateral damage from the AI arms race.
Oracle is locked into a $156 billion deal with OpenAI. With banks pulling back from financing data centers, Oracle’s borrowing costs have doubled and the stock has dropped. Payroll is the fastest place to cut. This isn’t AI replacing work; it’s overcommitting to the buildout and making people pay for the GPU bill.
The real lesson: betting on AI infrastructure as a moat without understanding the returns is dangerous. Oracle went all in to compete with AWS and Azure. The demand is there, and contracts are signed, but the timing and financing created a squeeze that lands on the workforce.
Advice:
Watch Oracle’s earnings this week. How they frame the cuts will reveal how companies talk about AI-driven restructuring. The framing will shape how boards and investors judge similar moves elsewhere.
That’s it for this week.
Two stories this week have a common theme: capital is concentrating, deals are fewer, and every hiring decision faces more scrutiny than ever. Anthropic shows which roles are at risk. LatAm VC data reveals the funding environment for builders in the region. Oracle shows what happens when you bet on AI before the economics make sense.
I help founders and operators build teams in Latin America. The most important question now isn’t how many people, but which people and why those roles matter right now. If you want to talk about it, you know where to find me: lupahire.com or just reply here.
Until next time,
Joseph Burns
CEO & Founder, Lupa