si querés entender cómo funciona un agente, construí uno desde cero.
usar Claude Code, Codex o Cursor te sirve para aprender a trabajar con agentes, pero no te muestra lo más importante: cómo funcionan por dentro.
empezá por las piezas que hacen que un agente funcione:
1) LLM
conectá un modelo. mandale una tarea simple y recibí una respuesta. esa es la unidad mínima:
input → modelo → output.
2) instrucciones
agregá un system prompt. ahí definís cómo tiene que comportarse el agente: qué rol cumple, qué reglas sigue, qué puede usar, qué debe evitar y cómo tiene que responder.
3) contexto
pasale la información que necesita para trabajar: historial, archivos, documentación, datos del usuario o estado del proyecto. el agente trabaja con lo que le das.
4) tools
sumale funciones que le permitan actuar: leer un archivo, buscar información, llamar una API o ejecutar un comando. el modelo no las ejecuta solo: las pide, y tu código las ejecuta.
5) memoria
guardá fuera del chat lo que conviene recordar: decisiones, preferencias, resultados o estado del proyecto. después lo recuperás y se lo volvés a pasar como contexto.
6) loop
un agente no responde una sola vez: recibe una tarea -> mira el contexto -> decide el próximo paso -> pide una tool si la necesita -> observa el resultado -> vuelve a decidir.
ese ciclo se repite hasta que puede responder, pedir más información o frenar porque no sabe cómo seguir.
construir uno desde cero te obliga a tocar cada pieza.
y ahí está el verdadero aprendizaje: entendés qué hace el modelo, cómo lo guían las instrucciones, por qué importa el contexto, cuándo usar tools, qué conviene guardar en memoria y cómo todo eso se conecta dentro de un loop.
después de eso, cualquier agente que uses se vuelve mucho más fácil de entender, diseñar y mejorar.
An entirely new network is forming:
The AI Filmmaker network.
We're not going to film school and meeting our collaborators anymore, cause film school is a scam and they hate AI for the most part.
These days, we go online and find the artists we like; we click like, we drop a comment, and give them a follow. And then sometimes we end up in a group chat, start talking, sharing works in progress, and then before you know it, we're collaborating.
This is the new film school. X is the new film school. Instagram is the new film school, and it's up to us to be proactive, honest, and open.
This is exactly how Eric (@gen_ericai) and I ended up collaborating on his incredible short film, 'All One'.
We clicked, connected, and collabed.
I say this to encourage you. We don't need expensive institutions to teach us, connect us, and empower us anymore. We just need to want it, and then go get it.
🔴 La IA genera texto pero no derechos de autor.
A Hernán Vega Posada le denegaron el registro de "su libro" por algo poco común: dijo la verdad.
En su solicitud ante el INDECOPI agradeció, sin más, a "ChatGPT-4o, quien fue el que escribió el libro". Con esa confesión, la Dirección de Derecho de Autor resolvió lo previsible (Resolución N.° 1111-2025/DDA): sin impronta humana —elecciones libres y creativas—, no hay originalidad; sin originalidad, no hay obra protegible. La IA no es autora.
La decisión es correcta. Pero deja al descubierto una paradoja: el sistema, tal como funciona hoy, castiga la transparencia. Quien hubiera ocultado el uso de IA y firmado el resultado como creación propia, probablemente habría obtenido su registro. Vega Posada perdió, en los hechos, por honesto.
Pero, ¿qué ocurre cuando alguien sí miente y presenta como propia una obra producida sustancialmente por IA?
Despejemos primero un falso problema. La IA no tiene derechos morales ni patrimoniales, no crea con intención y no puede ser agraviada. No existe una "víctima IA" a quien se le usurpe la autoría. Pero ausencia de víctima algorítmica no es ausencia de daño. El daño, cuando existe, es siempre contra personas. Y conviene nombrarlo con categorías jurídicas, sin emociones:
1️⃣ Contra quien contrató y pagó por talento humano y recibió output automatizado: estafa (art. 196 del Código Penal), si concurren engaño, error y disposición patrimonial; en sede civil, dolo como vicio del consentimiento (art. 210 del Código Civil).
2️⃣ Contra el mercado. Ofrecer como creación humana lo que fue producción de IA puede implicar competencia desleal por engaño (art. 8 del D. Leg. 1044). Detalle nada menor: vuelve a ser cancha del INDECOPI.
3️⃣ Contra el registro público. Aquí vale ser preciso: el registro de derecho de autor en el Perú es declarativo, no constitutivo; inscribir una obra no original no crea un derecho que no existe. Y la vía penal es cuesta arriba: por ejemplo, la falsedad ideológica (art. 428) solo alcanza al documento público y a un hecho que este deba probar, y la declaración del solicitante suele ser documento privado. No hay conclusión automática.
Queda el matiz decisivo, donde realmente se jugará esta disciplina: usar IA como herramienta no borra la autoría humana. Si una persona genera borradores, selecciona, reescribe, reestructura y transforma con decisiones creativas propias, esa capa humana puede ser protegible. Pero —y la propia resolución lo subraya— ni siquiera una contribución esencial vía prompts basta: se exige un "toque personal". ¿Dónde termina el prompt y empieza la obra? Es un debate aún abierto, no un asunto resuelto.
En todo caso, la regla general no cambia: usar IA no es ilícito. Mentir sobre su uso, o firmar como propio lo que no lo es, sí puede tener consecuencias jurídicas.
El derecho de autor del futuro no dependerá de saber si la IA puede "crear". La clave es distinguir cuándo el humano efectivamente creó y cuándo solo hizo clic.
Hace nada tenía 0 proyectos propios y un trabajo de oficina de 9 a 5, con 30 minutos de coche de ida y otros 30 de vuelta todos los días.
Hoy tengo varios proyectos activos a la vez y estoy a nada de empezar en una startup full remoto. Un gran cambio que hice fue lo que metía en mi cabeza cada día, y buena parte de eso fueron tres podcasts.
- Tengo un plan me encendió la parte más creativa, las ganas de emprender y de buscar otras formas de generar dinero. Traen a gente con unas historias que te abren la mente a un mundo que no sabías que existía.
- Finect Talk me mantiene actualizado y me hace entender cada vez más conceptos, por qué pasan ciertas cosas y no otras. Muchas veces acabo metiéndome a investigar un término por mi cuenta solo porque me picó la curiosidad.
- The Pragmatic Engineer me reactiva la chispa de desarrollador. Con esto y la IA tengo unas ganas de construir que no me las quito de encima, y escuchar a gente que trabaja dentro de las grandes me inspira a ser más.
Empezar a alimentar la cabeza con esta gente es lo que me empujó a abrir esto. A construir más proyectos, unos para clientes y otros de cara al público. A buscar trabajos nuevos. A desarrollar ideas que antes ni se me habían pasado por la cabeza. Y, sobre todo, a perderle el miedo a probar mil cosas aunque al final solo me quede con unas pocas.
un hack que uso cuando trabajo en tareas grandes con agentes:
les pido que mantengan un ADR (architecture decision record) con las decisiones que van tomando.
qué decidieron, por qué lo decidieron y si fue una instrucción mía o una suposición del agente.
después puedo usar ese archivo como documentación o dárselo a otro agente para que analice el trabajo sin tener que reconstruir todo el contexto desde cero.
The animated short film "Nube" (2023) is now online.
Directed by Christian Arredondo & Diego Sanchez de la Barquera.
The story of a cloud mother and her daughter.
>> https://t.co/qjsuHXWBcl
What Midjourney is:
- No investors, fully community-funded research lab
- Revenue from image generation product funds all R&D
- ~$100M in first 9 months, $200M by month 12, still growing
- 8 active projects: 4 hardware, 4 software
----2 hardware products coming to market soon (consumer-purchasable)
----2 are large-scale machines
DAVID HOLZ: Background and Philosophy
- Grew up in Florida, parents in medicine, dad had a dental office on a sailboat
- Physics and math background: drawn to the tension between predicting reality vs. absolute truth
- Core thesis: the interaction between humans and technology is the biggest limitation, not compute power
- Founded Leap Motion at 22: $10M in pre-orders in 48 hours from a website (not Kickstarter)
- Built hand-tracking VR: 600M-parameter mixture-of-experts model, 2015, CPU cluster, pre-TensorFlow
- Also shipped Northstar, an open-source AR headset
- Left Leap Motion wanting a “home,” not a 100x return
- Mentor Bill Warner told him he could bootstrap; he listened this time
- Started Midjourney with ~$200K, called Google for 10,000 GPUs on trust alone
THE SCANNER: Full Body Ultrasonic CT
- First new whole-body medical imaging modality in ~50 years
- Concept: “as powerful as an MRI, as casual as a trip to the spa”
- No radiation, no magnets, no x-rays; safe for unlimited scans
How it works:
- 40 rings, each with 8,960 transducers (200 microns wide), totaling 358,000 elements
- Fires ultrasonic waves at 100M times/second; sound travels through water at 1,481 m/s
- Sensors resolve motion down to picometers (sub-atomic range)
- Captures 17 GB/second of raw data; 806 TB per full scan reconstruction
- 21 on-site servers, 2 petaflops of compute
- Patient lowers into water at 4 cm/second; ~60 seconds for several hundred body slices
- Produces sub-millimeter 3D maps of internal tissue
Already outperforming MRI in some tissue boundary and muscle fiber detail on DAY ONE.
10x cheaper and 60x faster than MRI machines; scan cost effectively near zero.
Gen 2 scanner planned by end of 2026; Gen 3 will use custom silicon.
SCANNER vs. MRI: Key Differences
- MRI: 60-minute tube, loud, requires sedation for children, expensive, radiation-adjacent.
- This scanner: water immersion, 30-60 seconds, no sedation, no radiation, repeatable daily.
- Current limitation: not yet FDA-cleared beyond body composition; no AI layer yet applied.
- Already better than MRI in certain muscle/fiber/vein boundary resolution at day one
THE MIDJOURNEY SPA:
- First location: Union Square, San Francisco
- 25,000 sq ft, 4 floors
- Amenities: hot tubs, saunas, cold plunges, European spa features, gym
- 9-10 full scanners on-site
- Goal: open by end of 2027
Target:
- 50,000 scanners globally, capable of 1 billion scans/month
- 5,000 spa locations needed at ~10 scanners each
- Estimated $20B capex to scale; Midjourney self-funds the first location
- Payback period modeled at ~6 months per location
ROADMAP & REGULATORY PATH:
- FDA discussions already started; body composition on a clear path
- Ascending approval ladder planned:Body composition (near-term, easy)
- Sharing data with physicians
- Doppler / blood flow imaging
- Pregnancy / fetal imaging (ultrasound already approved; this is a natural extension)
- Therapeutic applications (tendon/muscle healing, eventually incisionless surgery)
AI not yet applied to imaging; planned as a layer once data volume grows
LONG-TERM VISION:
- Flag anomalies automatically, substitute some blood tests, enable daily health tracking
PRICING:
- No firm numbers yet; likely spa memberships plus walk-in and scan-only tiers; cost of scan itself is near zero
- Data analysis: day one is body composition only; physician sharing gated on FDA progression
- Form factors: current design is throughput-optimized (up/down elevator); bathtub and gym-sized variants possible later
- Blood test substitution: sub-millimeter daily differentials with AI may eventually replace some tests; acknowledged as frontier science
- Cancer destruction via focused ultrasound: technically possible, not on near-term roadmap
NEXT STEPS:
- Sign up for Midjourney Medical email list for research trial scan invitations
- Visit https://t.co/mWHl3Bj5WC for jobs and updates (page now live)
Gen 2 scanner presentation planned before end of 2026
More secret projects to be announced soon.
Más: gestor de tareas, temporizador Pomodoro, accesos directos y notas. Todo en un solo lugar. 100% gratis.
Lo hicimos para que ningún latino sienta que las becas y oportunidades son para otros. Instálala 👉 https://t.co/70mgASXpMX y dale RT para que llegue a todo LATAM.💜
Ayer vi Mandalorian and Grogu y para un fan real de Star Wars puedo decir que me emocionó, me intrigó y me sorprendió. Rotta y Grogu se llevan el crédito. Me sorprendió ver aMartin Scorsese en los créditos iniciales. 9/10
Liftoff of Starship V3, from the dunes right outside the pad.
This is the most insane shockwave action I have ever seen on video. Absolutely mad.
📽️ Me for @WeAreSpaceScout
Today we reduced headcount by 22%. The business is the strongest it's ever been. So I think it's important to be direct about what I'm seeing and why.
First, I made this decision and I own it. I did it because the way to operate at the highest level of productivity is changing, and to win the future, ClickUp needs to change with it.
Second, this wasn't about cutting costs. Most savings from this change will flow directly back into the people who stay. We'll be introducing million-dollar salary bands. If you create outsized impact using AI, you'll be paid outside of traditional bands.
Most importantly, I have the deepest gratitude for those affected. We're doing this from a position of strength specifically so we can take care of people properly. Everyone affected receives a package aimed at honoring their contributions and easing the transition.
I only see two options: wait for this to play out gradually in the market or be honest about what I'm seeing and act proactively.
THE 100X ORGANIZATION
The primary change is that we're restructuring around what I call 100x org. The goal is 100x output. The roles required to build at the highest level are fundamentally different than they were a year ago.
Incremental improvements to existing systems won't get us there. We need new ones. That means creating enough disruption to rebuild rather than iterate on what's already broken.
The common narrative is that AI makes everyone more productive. It doesn't. Many of the workflows of today, if left unchanged, create bottlenecks in AI systems.
These roles will evolve. But waiting for that to happen naturally means falling behind now.
The 100x org is actually heavily dependent on people - infinitely more than today. This is only possible with 10x people that have embraced and adopted new ways of working.
THE BUILDERS, AGENT MANAGERS, AND FRONT-LINERS
— THE BUILDERS: 10X ENGINEERS
I don't think most companies have internalized what's actually happening with AI in engineering. The common narrative is that AI makes all engineers more productive. That may be true in isolation, but at an organization level - that is the farthest thing from reality.
Here's what we've validated recently at ClickUp: the great engineers, the ones who can orchestrate, architect, and review, are becoming 100x engineers. They're not writing code. They're directing agents that write code. The skill is judgment.
AI makes the best engineers wildly more productive, and everyone else using AI slows these engineers down.
Think about it - the bottlenecks are (1) orchestration - telling AI what to do, and (2) reviewing - what AI did. Everything is leapfrogged and no longer needed.
So who do you want orchestrating and reviewing code?
And how do you want your best engineers to spend their time?
If your best engineers are spending time reviewing other people's code, then this is inherently an inefficient bottleneck. These engineers can review their agent's code much faster than reviewing human code.
The new world is about enabling your 10x engineers to become 100x.
The wrong strategy is to push every engineer to use infinite tokens. Companies doing this are celebrating 500% more pull requests. But customer outcomes don't match the volume of code being generated.
I call this the great reckoning of AI coding, and every company will face this soon if not already.
More code is just another bottleneck to the best engineers, and ultimately to your company's impact as well.
— THE BUILDERS: 10X PRODUCT MANAGERS
Product management and design roles are merging.
Designers that have customer focus, become more like product managers.
And product managers that have intuition for UX become more like designers.
The bottleneck of user research is gone. It takes us just one mention of an agent to kickoff research and analyze results.
The bottleneck of product <> design iteration is also gone. The product builder iterates on their own, along with agents and skills that ensure alignment with quality and strategy.
Also controversial today - I believe that the wrong strategy is to have your PMs shipping code - that just introduces another bottleneck that the best engineers will waste their time on.
To be clear, PMs should be coding but they should do this in a playground to iterate, validate, and scope. That code should not go to production.
Everything outside of managing systems, orchestrating AI, and reviewing output becomes a bottleneck.
That's why the other roles that are critical along with these are the systems managers (to reduce bottlenecks) along with a bottleneck you can't replace - customer meeting time.
— THE SYSTEM MANAGERS
Ironically, the people that automate their jobs with AI will always have a job. They become owners of the AI systems - agent managers. We have many examples of these people at ClickUp.
The underlying systems in which we operate are absolutely critical to get right. I think most companies are delusional to think they can iterate on existing systems and compete in this new world.
You must create enough disruption so that old systems are deprecated entirely. If there's any definition for 'AI native' that's what it is.
— THE FRONT-LINERS
In a world that will become saturated with AI communication, the human touch will matter more than anything to customers.
This is a bottleneck that you shouldn't replace - even when agents are high enough quality to do video meetings.
One-on-one meeting time with customers is something that shouldn't be automated. The systems around the meetings should be - so that front-liners spend nearly 100% of their time with customers.
REWARDING 100X IMPACT
In a world where companies are able to do so much more with less, where does that excess money go?
In our case, much of the savings in this new operating model will flow directly back to those that enabled it.
We must reward people that create productivity accordingly. This aligns incentives on both sides. Plus, in a world where your best people create 100x impact, you can't afford to lose them.
You should aim to retain these employees for decades. The context they have and their ability to efficiently orchestrate and review will be nearly impossible to replace.
Compensation bands of today should be thrown out the door. We're introducing $1 million cash/year salary bands with a path available to nearly everyone in the company if they produce 100x impact by creating or managing AI systems.
THE FUTURE
Nearly every company will make changes like these. The ones that do it proactively will define what comes next.
The future is not fewer people. It's different work, new roles, and better rewards for those who embrace it. We're already seeing entirely new roles emerge, like Agent Managers, that didn't exist a year ago.
ClickUp is positioning to lead this shift, not just internally, but for our customers too. I've never been more certain about where we're headed.