In 2016 I didn’t know what a server was. Ten years later I’d directed a Fortnite island 65,000 people played at the same time.
For a decade I directed live products for the biggest Spanish-speaking creators — series, servers, islands, economies — before I knew the job had a name. Today I build multi-agent AI systems under the same rule I learned there: the deadline is part of the product.
65,000Peak concurrent players on one island — #1 across all of Fortnite
5–6 / 20+UEFN islands directed end to end / contributed to in specialist roles
10+Commissioned live series & events directed over 10 years
577KSubscribers on my own YouTube channel
The work 2016 — today
These tracks overlap. I ran commissioned series, my own network and UEFN islands at the same time, not one after another.
012016Server admin · Series director
The game had to fit the format, so I changed the game
My first series was a live ARK: Survival Evolved run on my own channel. On paper the job was infrastructure: stand the servers up, configure them, keep them alive, administer the series day to day.
What the series actually needed was something else. An episode has to end somewhere, and the game’s default tuning has no opinion about episodes. So I stopped treating the configuration as a technical constant and started treating it as a design surface — tuned so that every recording session produced an episode worth publishing.
I had no vocabulary for it at the time. It was the first product decision I ever made.
Server admin · Series directorMy own ARK series, episode 1 — Lexosi Plays · February 2016
022016 — todayServer admin · Series director
They kept commissioning me across games because what traveled was the execution
Ten-plus commissioned live-series and event productions over ten years, starting with ARK in 2016 and expanding — at the creators’ request — to GTA V, Minecraft, Rust and more, for Rubius, Vegetta777, Willyrex, Alexby11, xFarganx, sTaXx, Los Noobs and others.
On every one of them I was both the server admin and the director of the series. The creator would tell me what they wanted for the next episode, and my job was to make it possible: change the server, build whatever was missing, decide what it cost. When what they asked for couldn’t be done, I said so — not possible — and we found something else that gave them just as good an episode. And often the question came the other way round: we’ve already shot this one, what do we do next? That’s when the one proposing was me.
The rest was the same every time, and it was never really about the game: research, assembling the team, budgeting, game systems and world rules, daily content design so each creator had fresh material every episode, live operations on running servers, and integrated in-game purchase systems in several productions.
On Piratas we shipped physics Minecraft didn’t have: players built their own ships and sailed them. Game-agnostic by track record, not by claim — and all of it is public.
The series you can watch here run to 2021. The track didn’t stop there — the recent productions are under NDA, and that is the only reason they aren’t on this page.
I told this story on my own channel in June 2021, five years before any of it was a CV.
In my own words · 2021How Lexosi met Vegetta777 and Willyrex — Lexosi Plays · in Spanish
Every technical person Willyrex had worked with said yes to everything and then missed the deadline. I did the opposite: I reformulated the request into what he actually needed on screen and delivered it the next day, instead of building the literal version over weeks.
The deadline is part of the product.
032020 — 2021Founder · Operator
Tomatecraft: the decision that made it work was letting the two halves play together
In 2020 I founded and ran my own Minecraft network. The thing that made it work wasn’t a feature, it was a platform call: Java players on PC and mobile players in the same world, at a time when almost no server offered crossplay.
It was self-funded out of my YouTube income, with programmers, designers and builders on payroll. That changes how you decide. Every feature I cut, I cut knowing what the payroll cost that month.
A year and a bit later we merged Tomatecraft with Noobcraft, Arsilex’s network, and ran large branded live events on top of them.
Network owner · Live opsLive event on my own network — 100 subscribers, Vegetta777 and Arsilex · 2021
042022 — 2026Directed / Collaborated
Five to six islands end to end — and the pricing on twenty more
Game designer and de-facto PM of 5–6 complete Fortnite (UEFN) islands, end to end: market scouting — identifying game concepts proven on Roblox and adapting them to Fortnite’s audience and platform constraints — mechanics and mode design, economy and progression, playtesting, and live-ops iteration on real player data, leading a core team of 5.
Beyond the ones I directed, I contributed to 20+ other islands in specialist roles: gameplay programming in Verse, configuration, 3D modeling, mechanics design, and — most often — pricing strategy for integrated V-Bucks micro-purchases.
Some of these I directed; on others I did Verse, 3D modeling, level design, configuration or pricing. Small teams — everyone did everything. I don’t put a job title next to each one because it would be inventing a precision that never existed.
The hardest purchase is the first one. After that, buying again feels natural.
So I priced the first one to be bought.
05August 2024Directed · Core team of 5
65,000 at once, ahead of Epic’s own maps
The Fall Guys-collab island — ONLYUP FALL GUYS! — was conceived and run under that method. It peaked at 65,000 concurrent players.
For several hours on day 2 of the collaboration it was the #1 most-played island globally across all of Fortnite. It held #1 in the Fall Guys category for two weeks — ahead of Epic’s own official maps for the event.
I directed it. I didn’t do it alone: core team of 5.
Directed · Core team of 5Vegetta777 & Willyrex playing the island live · August 2024
The last thing I shipped there was a system, not a map
A battle-pass system for a UEFN island with direct in-island monetization via V-Bucks — VIP pass, level purchases and score multipliers — built from the start as a reusable framework rather than a one-off.
By the time I left it had been adopted across 4 maps. It was my final project with the studio.
What shipped
VIP pass
Level purchases
Score multipliers
Maps that adopted it4
I’ve also advised on social media strategy for public institutions and large organisations. Those engagements are under NDA.
Then I did the same thing to my own tools 2026 — today
Rules an agent can weigh get weighed away
I built a multi-agent orchestration system on my own: 19 agents — 13 Claude reasoners and 6 DeepSeek workers — root-driven, with human stop-gates and a full audit trail.
I wrote the rules as advisory prose first. Measured, they produced 0/3 compliance in the three runs right after I canonized them: the hook fired, the agent proceeded anyway, and justified itself on the way past. So I moved enforcement out of the prose and into the tool-call layer, deny-by-default. What the logs show since:
all 9 blocked tool-calls across 7 runs, 4 via explicit logged override, 0 unauthorized writes
The reliability layer is extracted and public as loopward — anti-loop, human stop-gates, audit trail — pip-installable, offline demo, CI-verified. The measurement is written up in full, with the numbers reproducible by script.