Skip to slide 1
01 / 20
VCN LATAM · LinkedIn Live · 16 JUL 2026 · SF 4:30 PM · BOG 6:30 PM
● en vivo

prompts que escriben prompts / con Claude Code / español + english

Irina Vélez · Colombia  ·  Rayyan Zahid · San Francisco
> archive_log: open
> loading prompt_compiler
● 200 OK · listo
ACT I    this talk is built from receipts

0
of our own Claude Code sessions,
mined for patterns.

→ data/prompt_library/_corpus · 403 human prompts · 5,963 machine calls · sesiones reales, no teoría

el problema real

You know what you want. You don't know how to ask for it.

Sabes lo que quieres. No sabes cómo pedírselo a la IA. Ahí entra el meta prompting.

definición · four shapes, one word

META PROMPTING IS FOUR MOVES.

01
WRITE. One model call generates the prompt another call will follow. La IA escribe el prompt por ti.
02
EVALUATE. A model scores or critiques a prompt against a rubric. La IA califica el prompt.
03
LOOP. Generate, evaluate, improve. Repeat until it measurably wins. El bucle: generar, evaluar, mejorar.
04
CONDUCT. One conductor model spins up expert calls of itself. Claude Code subagents are this, in production. Un director, muchos expertos.
la única línea que hay que recordar

Stop writing the prompt. Write the thing that writes the prompt.

Deja de escribir el prompt. Escribe lo que escribe el prompt.

el mecanismo detrás de todo lo que sigue

ONE LOOP. EVERY TECHNIQUE ON THE LADDER.

rough ask pregunta cruda score each lap mantén lo que gana after N laps 01 · DRAFT genera un borrador 02 · CRITIQUE el modelo se evalúa 03 · REWRITE reescribe con la crítica CONVERGED PROMPT kept by score, not vibes elegido por puntaje

→ Self-Refine, OPRO, y GEPA corren este mismo bucle · solo cambia quién juzga la crítica

ACT II    the research ladder · 2022 to 2025

SEVEN RUNGS. EACH ONE MEASURED.

nueve papers leídos hoy · los PDFs están en el repo de la charla

ACT II    la escalera · every rung, mapped

THE LADDER. CLIMB IT IN ORDER.

  1. 07
    GEPA reflection beats reinforcement Agrawal et al. 2025
  2. 06
    TextGrad gradients are sentences Yuksekgonul et al. 2024
  3. 05
    DSPy declare signatures, compile Khattab et al. 2023
  4. 04
    Promptbreeder mutators also evolve Fernando et al. 2023
  5. 03
    OPRO history becomes prompt Yang et al. 2023
  6. 02
    Self-Refine generate, critique, rewrite Madaan et al. 2023
  7. 01
    APE model writes instructions Zhou et al. 2022

→ un peldaño no reemplaza al anterior. cada uno se apoya en el que sigue.

rungs 01 + 02 · the loop you can build tonight

A MODEL THAT CRITIQUES ITSELF ALREADY WINS.

SELF-REFINE
~20% average gain over single-shot output across 7 tasks. Three prompts: generate, critique, revise. No training, no infra. Madaan et al. 2023.
APE
24 of 24 tasks matched or beat human-written instructions. Machine IQM 0.810 vs human 0.749. Zhou et al. 2022. El modelo ya escribe instrucciones a nivel humano.
rung 03 · OPRO · Google DeepMind

"TAKE A DEEP BREATH"
scored 80.2% on GSM8K math.
the human classic scored 71.8%.

→ the model found that prompt by reading its own score history. nobody wrote it. · Yang et al. 2023 · PaLM 2-L · no-instruction baseline: 34.0%

rung 04 · evolution eats the evolver

PROMPTBREEDER MUTATES THE MUTATOR.

A genetic algorithm over a population of prompts. The twist: the mutation operators are themselves prompts, and they also get mutated and selected. Self-referential all the way down. La evolución también evoluciona.

58.4%
hand-written Chain of ThoughtGSM8K zero-shot · PaLM 2-L
80.2%
OPRO discovered promptoptimization by prompting
83.9%
Promptbreeder evolved promptFernando et al. 2023
rungs 05 + 06 · compilers and gradients

BACKPROP WORKED FOR WEIGHTS. NOW IT WORKS FOR WORDS.

DSPY
33% to 82% on a GPT-3.5 math pipeline, with zero hand-written prompts. You declare input and output signatures; a compiler bootstraps the demos and optimizes the pipeline. Khattab et al. 2023.
TEXTGRAD
GPQA 51% to 55% on GPT-4o, plus a 20% relative gain on LeetCode Hard. Every component is a graph node; natural-language "gradients" flow backward telling each node how to change. Yuksekgonul et al. 2024. Los gradientes son oraciones.
rung 07 · 2025 · the frontier

GEPA: REFLECTION BEATS REINFORCEMENT LEARNING.

+20%
over GRPO reinforcement learning, at best+6% on average · Agrawal et al. 2025 · ICLR 2026 oral
35x
fewer rollouts than the RL it beatlanguage feedback is denser than a scalar reward
+10%
over MIPROv2, the prior best prompt optimizerincluding +12% on AIME 2025 math

→ reflexiona en lenguaje natural sobre sus propios trazos de ejecución, muta el prompt, y guarda una frontera de Pareto de ganadores diversos. No necesitas RL. Necesitas un buen bucle.

rung 07, en detalle · words vs. weights

WORDS BEAT WEIGHTS. SAME RESULT, FEWER TRIES.

RL pipeline · GRPO 24,000 rollouts

Thousands of gradient rollouts to fit one task. Miles de intentos, una señal numérica cada vez.

GEPA · reflective evolution up to 35× fewer

Reflect on the trace, mutate the prompt, keep the Pareto winners. Reflexionar, mutar, seleccionar. Eso es todo.

→ GEPA beats GRPO by 6% on average, up to 20%, using up to 35x fewer rollouts. Agrawal et al. 2025, ICLR 2026 oral. El lenguaje es un medio de aprendizaje más rico que un premio escalar.

rung 07, in plain english · how it learns

IT READS ITS OWN FAILURES. THEN REWRITES ITS OWN INSTRUCTIONS.

→ your error messages are training data. Tus mensajes de error son datos de entrenamiento. Un número te dice que fallaste; el lenguaje te dice por qué.

EL BUCLE
SE CIERRA.

The machine now searches prompt space better than you do. The question is no longer how to write a prompt. It is what loop you hand your criterion to.

ACT III    se lo hicimos a nuestro propio repo

THE PLAYBOOK, IN PRODUCTION.

paso uno · mine your own history

YOUR HISTORY IS TRAINING DATA.

6,366
total Claude Code sessions on disk1.18 GB of JSONL transcripts
5,963
machine sessions, filtered outautomation calling claude -p under the hood
403
prompts a human actually typedthe only ones that teach you how you really ask
18
named, parameterized pattern templatesplus a 10-entry directive vocabulary, each with an evidence trail

→ tu historial ya es un dataset. Los prompts que repites son plantillas esperando nombre.

paso dos · route, honestly measured

CHEAPEST TIER FIRST. MEASURED, NOT VIBES.

Alias hit, then keyword signature, then an LLM for the long tail. The signatures live once, in the pattern file. Tres niveles: alias, keyword, LLM.

56%
of 397 real prompts claimed by keyword signaturesthe other 43% is a genuine long tail. it goes to the LLM tier
36/36
P02 improve-after-build routing precisiona strong keyword signal earns its fast path
0/8
P06 events-workflow keyword precisiontoo broad. it should always fall through to the LLM. knowing this is the point
demo en vivo · recorded output as the safety net

ONE TERSE ASK BECOMES THE FULL PROMPT.

terminal · recorded 22 minutes before this talk prompts.route
$ python main.py prompts route "reverse https://sponsor-dashboard.example.com" --no-llm
# pattern: reverse-api (P04) | high via keyword
# dispatch -> skill:reverse-api / agent:reverse-api-builder
Reverse https://sponsor-dashboard.example.com. I'll help with login.
Follow our 4-layer pattern: CDP bootstrap -> curl_cffi client ->
kernel verbs -> CLI. Document the captured endpoints, use typed
errors, and smoke-test one low-stakes write before any bulk operation.
Directives:
- Smoke / e2e test before declaring done. Build-green is not correctness.

→ also live on stage: python main.py prompts list · prompts directives · runnable mini version: vcn-latam-metaprompting.vercel.app/examples/06_router.py

Suzgun & Kalai 2024 · one LM, many fresh instances

ONE CONDUCTOR. FRESH EXPERTS EVERY CALL.

The conductor decomposes the task, writes each expert's instruction, and calls a clean instance for every one. No expert sees the task, the transcript, or the other experts — only its own brief. Un director, expertos frescos: cada uno ve solo su instrucción.

CONDUCTOR decompose · verify MATHEMATICIAN fresh instance CRITIC fresh instance DOMAIN EXPERT fresh instance PYTHON TOOL fresh instance

→ experts chosen per task, briefs written by the conductor · synthesize + flag contradictions · runnable tonight: /examples/04_conductor.py

llevatelos a casa · the take-home pack

SIX SCRIPTS. ZERO API KEYS.

01
$ python 01_self_refine.py "your task"generate, critique, revise · el bucle de esta noche · Self-Refine 2023
02
$ python 02_ape.pythe model reverse-engineers your instruction from examples · APE 2022
03
$ python 03_opro.pythe score history becomes the next prompt · OPRO 2023
04
$ python 04_conductor.py "your question"one conductor, fresh expert instances · Meta-Prompting 2024
05
$ python 05_criterion.pyyour judgment, versioned + cached · run it twice and watch the cache
06
$ python 06_router.py "your ask"3-tier template router · no model call at all

todo en vcn-latam-metaprompting.vercel.app/examples/ · pure stdlib · runs on the Claude Code CLI you already have

ACT IV    proof, not promise
Exhibit

This
deckbuilt by a metaprompt · today · in one session

The prompt that built this deck ships inside the deck. Fetch it yourself, right now.

  • Moodboard: a Pinterest board, reduced to a palette by a script
  • Research: two agents read 9 papers and mined 1 repo
  • Copy + build: one conductor session, CC-PRO method
  • El prompt que construyó este deck viaja dentro del deck
terminal · live awaiting
> curl /.well-known/ai-agent.json
one rung up · and one honest confession

AGENTS THAT EDIT THE AGENTS' INSTRUCTIONS.

CALIBRATION
Reads session telemetry, finds what an agent keeps getting wrong, edits the agent's instruction file. Evidence-gated. Never touches working code. Un agente que corrige el manual de otro agente.
OUROBOROS
Plan, critique through adversarial lenses, revise, K layers, stop on convergence. Its own docs name the lineage: Self-Refine, Reflexion, Debate, CRITIC. Honestidad: ingeniería bien hecha, no un mecanismo nuevo.
THE FRAME
Most "new" agent frameworks are one of these loops wearing a trench coat. Knowing the ladder means you can name what you're actually running. Y saber qué estás corriendo es la mitad del criterio.
CALIBRATION    el manual, no el código

THE AGENT THAT REWRITES THE MANUAL.

Same mistake, three sessions running — then the instruction file changes. Nunca el código de trabajo.

6,366 sessions 6,366 sesiones NEXT RUN próxima corrida TELEMETRY stack of run logs PATTERN same mistake × 3 EVIDENCE GATE EDIT agent.md − old line + new line

→ evidence-gated · never touches working code · un agente que corrige el manual de otro

el plan que se corrige a sí mismo · four lenses, three layers

OUROBOROS. PLAN, CRITIQUE, REPEAT.

01 02 03 PLAN v0 converged Δ→0 · stop skeptic user ops consequences Self-Refine Reflexion Debate CRITIC OUROBOROS

→ Self-Refine, Reflexion, Debate, CRITIC corren este mismo bucle · ingeniería bien hecha, no un mecanismo nuevo

ACT IV    abre el saco · name what you're running

OPEN THE COAT. NAME THE LOOP.

the hype stack the ladder · la escalera

→ Most "new" agent frameworks are one of the ladder's loops, renamed. Y saber qué estás corriendo es la mitad del criterio.

la frontera, este mes · six papers, ten months

THE FRONTIER, THIS MONTH.

01
The Optimizer That Remembers Its Own Training
SePO trains one agent to evolve both your prompt and its own optimizing skill together, keeping past attempts as stepping stones instead of throwing them away.
arXiv 2606.04465 · Jun 2026
02
Your Context Window Is Also a Prompt
SCOPE evolves several candidate prompts at once from an agent's own execution traces, routing lessons to quick fixes or long-term memory. Nearly 3x'd success on a hard benchmark.
arXiv 2512.15374 · Dec 2025 (v2 May 2026)
03
Give the Optimizer a Debugger
SPEAR lets the optimizer run actual code to cluster its own errors before rewriting a prompt, with automatic rollback the moment a change makes things worse.
arXiv 2605.26275 · May 2026
04
Two Memories, Not One Fresh Start
MemAPO keeps a running memory of what worked and what keeps failing, composing new prompts from both instead of re-optimizing every task from zero.
arXiv 2603.21520 · Mar 2026
05
More Data Can Make You Worse
p1 shows most training examples just measure the model's own noise, not whether a better prompt helps — keep only the few examples that truly discriminate, and beat the field with two.
arXiv 2604.08801 · Apr 2026

seis meses de arXiv, reducidos a cinco ideas · un builder las entiende en cinco segundos cada una

clónalos esta noche · the repos, not the theory

CLONE THESE TONIGHT.

repo stars what it does
36.2k
Programs LLM pipelines and auto-tunes prompts; hosts the GEPA optimizer.
23.3k
Eval and test harness for comparing and scoring prompts and agents.
4.2k
PyTorch-style trainer for LLM apps that auto-optimize their own prompts.
3.7k
Backpropagation through text feedback to auto-rewrite prompts.
764
Official OPRO code: the LLM-as-optimizer that opened tonight's talk.
758
Structure-aware, multi-objective metaprompt optimization library.
421
The original meta-prompting paper's code: one conductor, fresh experts.
181
A LangChain build of Promptbreeder — the mutator that mutates itself.

todos verificados hoy en GitHub · star counts as of tonight · graduate to these when the six starter scripts stop being enough

ACT V    empieza esta semana

THREE MOVES. ZERO INFRA.

01
> Self-Refine tonight: generate, critique, revise. Three prompts, any model.esta noche · ~20% average gain in the paper · no infrastructure, just the loop
02
> Version your criterion: put your judgment in a prompt file with a version number and a cache.tu criterio, versionado · bump the version, the cache invalidates itself
03
> Mine your own history: the prompts you keep retyping are templates waiting for a name.mina tu historial · nuestra versión: 6,366 sesiones a 18 plantillas con slots

all six examples, runnable tonight, zero API keys: vcn-latam-metaprompting.vercel.app/examples/ · graduate to DSPy or GEPA when you have a real metric. no antes.

cierre

El acceso al frontier no debería depender de tu código postal.

próxima sesión LATAM: fin de julio · las herramientas son las mismas en SF, Bogotá o CDMX. el playbook, ahora también.