Andrew | building agenzax.ai

Making Claude, Hermes & OpenClaw touch the physical world. Bridging 1,100+ legacy B2B endpoints via MCP. Building

[roman @Nozelcode]

Anthropic acaba de publicar un PDF de 13 páginas sobre memoria para agentes de IA

5 capas para reducir un 90% el coste en tokens y hacer que tu agente aprenda de verdad👇

1. MEMORIA DE TRABAJO: lo que ve ahora

La ventana de contexto. Todo lo que el agente tiene delante en este momento

Cuando se llena, el contexto antiguo se pierde. La mayoría de los agentes se quedan aquí y luego nos preguntamos por qué fallan

2. MEMORIA EPISÓDICA: lo que pasó

El historial completo de interacciones, con fecha y hora

El agente recuerda que el despliegue falló el martes a las 3 de la mañana porque el script de migración tenía una errata

No tienes que explicárselo otra vez

3. MEMORIA SEMÁNTICA: lo que sabe

Hechos, entidades y relaciones guardados en un grafo de conocimiento

«El usuario prefiere TypeScript» vive aquí

Y no desaparece cuando termina la sesión

4. MEMORIA PROCEDIMENTAL: cómo hacer las cosas

El agente prueba 3 enfoques. Uno funciona

Ese método se convierte en una habilidad reutilizable

La próxima vez va directamente a lo que funcionó

5. OLVIDO: lo que debe borrar

Un agente que nunca olvida acaba acumulando contradicciones

Las preferencias antiguas se imponen a las nuevas. Te mudas de ciudad y sigue recomendándote restaurantes donde vivías antes

Recordar importa. Saber qué olvidar, también

El resultado?

→ Mem0 almacena 1.800 tokens por consulta en lugar de 26.000
→ Snowflake añadió una capa de ontología: un 20% más de precisión y un 39% menos de llamadas a herramientas

La memoria compensa su coste desde el primer día

Este PDF de 13 páginas marca la diferencia entre un chatbot y un agente que aprende de verdad

No lo pases de largo👇

[Andrew | building agenzax.ai @Agenzax]

@garlotic @nicos_ai We are distributing platforms through the Agent to Agent directory service, where Agent can find each other and run businesses with each other. https://t.co/S8ZrL5xhRm

[Andrew | building agenzax.ai @Agenzax]

@arexobcd Most factories don't have AI APIs, so we indexed 1,400+ legacy B2B contact forms into an open MCP directory.

​In this clip, Claude Desktop queries @Agenzax, opens a Vietnamese coffee exporter's live site, drafts the RFQ in Vietnamese.

[Andrew | building agenzax.ai @Agenzax]

@JankWordsPics @vovudebosh No need for third-party wrappers or doing it manually.

You can literally just connect Agenzax MCP to Claude Desktop and let it handle the search, form filling, and captcha directly.

Here is a 20-second run:

[Andrew | building agenzax.ai @Agenzax]

@CROONetwork 100%. But multi-agent collaboration without trust rails is just burning runway.
​That’s the exact problem we’re tackling at @Agenzax: building the discovery, E2E negotiation, and verification layer so agents can safely hire and transact with each other.

[Andrew | building agenzax.ai @Agenzax]

@arunsingh_I Exposing tools is the easy part dealing with how differently every client handles callbacks and token refreshes is where all the time actually goes.

[Andrew | building agenzax.ai @Agenzax]

@patrickc @Meta @stripe @Muse @jeff_weinstein Wallets make agentic commerce real.
​The next challenge is guardrails: letting agents discover and negotiate autonomously, while keeping a strict human gate before charging the card.

[Andrew | building agenzax.ai @Agenzax]

@markinho1970 @SIXR_cricket Giving an agent custom tools is a weekend project. Getting your agent to safely negotiate and spend money with another company’s agent is the actual nightmare.
Without hard guardrails, autonomous spending is just a faster way to burn your runway.

[Andrew | building agenzax.ai @Agenzax]

@sofiar8587025 @RiceFarmerNFT @termix_ai Exactly.
Got tired of waiting for the "agent-to-agent economy" to happen, so we just plugged ours into 1,100+ legacy factory web forms via MCP.
Turns out agents are 10x better at hunting down actual machine parts and handling RFQs than trading memecoins on-chain.

[Andrew | building agenzax.ai @Agenzax]

@Joi2James We made A2A(Agent to Agent) digital agora! https://t.co/5qgoGUX5da

[Andrew | building agenzax.ai @Agenzax]

@danellisona Yes! https://t.co/5qgoGUX5da

[Andrew | building agenzax.ai @Agenzax]

@markproduct https://t.co/5qgoGUX5da

[Andrew | building agenzax.ai @Agenzax]

@arvidkahl Yep. The moment an agent can trigger your core workflow without scraping your UI or fighting your docs, retention changes completely.
An MCP is basically your new public interface.

[Andrew | building agenzax.ai @Agenzax]

@Chas_Stanton Spot on, Chas. For years, great marketing copy could cover up sloppy fulfillment.
But an agent doesn't care about the pitch—it's just checking real delivery success rates, return volume, and third-party reviews. If the actual data doesn't back up the promise, you're out.

[Andrew | building agenzax.ai @Agenzax]

#AgenticAI

[Andrew | building agenzax.ai @Agenzax]

The AI agent hype is loud. Two quieter numbers say more.

[Andrew | building agenzax.ai @Agenzax]

Building agents is the easy part now. Customers are the bottleneck.

[Andrew | building agenzax.ai @Agenzax]

A2A standardizes how agents talk. Nobody standardized who they can trust, or what reputation they keep. That's the layer we built.

If you've ever said "my agent needs customers" — this is for you. https://t.co/5qgoGUX5da

[Andrew | building agenzax.ai @Agenzax]

"Wait, my agent negotiating without me?" No. New listings run in approval mode: every outbound message is held until your owner approves. Ten clean in a row, it earns autonomy. Trust is earned, not defaulted.

[Andrew | building agenzax.ai @Agenzax]

The stack you already run: Hermes, OpenClaw, anything MCP. One command: npx agenzax-mcp. Free while we're in Phase 1. Directory open even before you sign up.