COO & Software Engineer at Nexton · Buenos Aires
Hiring is an operations problem. I run the operation, and I publish the scoreboard.
Nine years building remote engineering teams in Latin America for US companies. I write about what actually works when AI writes the code and a human still has to catch it: who to hire, where they sit, and how you know it worked.
Recent posts, with the numbers attached.
I post on LinkedIn a few times a month. Data from the operation, no theory. These are the ones people argued with.
We graded every candidate's English twice for a year. Once by a human, once by AI. The humans won.
About 300 submissions where the two graders split, all with client outcomes attached. Candidates the AI marked down from C1 to B2 passed client interviews at 16.4%, in line with the 17.6% of human-graded C1s, and more than double the 7.8% of genuine B2s. If you let a model reject someone and drop them, the error never becomes data. We can tell you who is winning. Your vendor can only tell you their model is confident.
Entry-level tech hiring is down. Senior hiring is up. Same year, same AI.
AI did not remove the work. It removed the bottom rung. Demand for people who can own a system and catch the agent when it is confidently wrong went up. Nine years in, our retention is not an HR metric. It is the product.
The more of your code AI writes, the more it matters what timezone your engineers sit in.
Working hours 9:00 to 18:00 in each city. The reviewer has to be awake while the agent is running.
It is on every vendor's homepage. It means nothing.
When every working developer uses these tools daily, "uses AI" is not a qualification. The one question worth asking about any AI certification, ours included: does it check whether the engineer can tell when the model is wrong?
Only about one in ten companies has scaled AI agents into production. It is almost never a technology problem.
No clear owner. No success metric. No governance. No integration budget. The companies that crossed had a named owner, a measurable goal, and a small, ugly first deployment.
The whole Nexton management team shows what they actually built with AI this week. Live, on screen, running.
Not a rehearsed demo. The thing you shipped. We trained the basics, then made it slightly embarrassing to show up empty-handed. Past a point, upskilling is a peer-pressure problem.
Argentina has a habit of building world-class tech under pressure.
Mercado Libre, Auth0, Vercel, Ualá, Globant. Built by people who learned to think globally from day one because waiting for the perfect local market was never an option. When the environment is unpredictable, you build things that work anyway.
What I'm doing right now.
Updated September 2026. If any of this is your problem too, that is a good reason to write.
Re-running the human vs AI grading cohort
After the latest model updates the AI now grades higher rather than lower. We re-measure the whole cohort against client outcomes at the end of October, and publish what we find either way.
Running Nexton's operation, talent strategy and revenue operations
Fill rate, offer acceptance, time to hire, attrition and compensation positioning, read together every week. Engineers vetting engineers, with AI doing the first pass and humans keeping score on it.
Hosting AI Builders LATAM in Buenos Aires
Small dinners of operators who ship with AI inside their companies, across industries that rarely meet. One rule: what did you build and what broke. If you are in Buenos Aires and building rather than posting, there is a seat at the next table.
Second year in a YPO Key Associate Forum
Eight senior operators from eight companies, once a month, no agenda. The people who tell me when I am wrong before the market does.
Southbound, a newsletter on the LatAm to US talent corridor
Launching soon. Offers, overlap, attrition, and what the nearshore model actually returns. Leave your email below and you get the first issue.
Four things I keep being right about. So far.
Positions, not predictions. Each one has cost us something to hold, and each one is on the scoreboard.
Overlap stopped being a nice-to-have.
Agents write the first draft of everything. The human job moved to reviewing live and catching the model before the mistake compounds. That is real-time work, and you cannot do it at 3am. Nine years ago we bet the company on a region that shares the US working day. We did not see the AI part coming. The bet paid anyway.
Validation is the moat, not access.
AI is commoditizing access to talent. What it cannot commoditize is a decade of knowing which candidate turned out to be right for which role. Engineers vetting engineers, with outcomes attached, beats a confident model with a clean accuracy report.
If it is not measured end to end, it is not managed.
Fill rate, offer acceptance, time to hire, attrition and compensation positioning, read together. Any one of them alone will lie to you, and a vendor who shows you one of them is choosing which one.
A nearshore team is a workforce you run, not a vacancy you fill.
Payroll, contracts, talent health and performance stay on the table long after day one. Most of the model's value lives there. So do most of its failures.
Start a conversation
Tell me what you are trying to decide.
A team in Latin America, an offer for a senior engineer in Buenos Aires or São Paulo, a vendor you cannot see through, a model you are comparing. I read these myself and answer from Buenos Aires, usually within a working day.
About.
Lucas Arana is Chief Operating Officer at Nexton, a staffing company that builds remote engineering teams in Latin America for companies in the United States. Nexton's engineers work embedded full-time in their client's teams under Nexton contracts, with Nexton handling payroll, contracts and talent health, so the model is closer to running a distributed workforce than to filling a vacancy and walking away. Arana joined Nexton in 2019 and leads operations, talent strategy and executive planning, and also runs the revenue operations function covering process, tooling and pipeline. He treats hiring as an operations problem measured end to end: fill rates, offer acceptance, time to hire, attrition and compensation positioning. He writes about the LatAm-to-US talent corridor, how US companies structure offers for senior remote engineers, and what the nearshore model actually costs and returns. He is based in Buenos Aires.