Lucas Bolle | mibo-ai.com

Building 🤖 | Ayudando a devs y no tan devs a testear sus agentes de IA sin que alucinen en producción. Automatización, Flowise, N8n y SaaS.

[Lucas Bolle | mibo-ai.com @lubolle]

@imryanarya007 https://t.co/KQanbBXsm8

[Lucas Bolle | mibo-ai.com @lubolle]

@adhamuxi https://t.co/TxMxZ2Sj2K

[Lucas Bolle | mibo-ai.com @lubolle]

We built Mibo’s passive evaluation entirely on top of it. No re running LLM calls, no sitting in the request path, just real visibility into what your agent actually did.

[Lucas Bolle | mibo-ai.com @lubolle]

AI agents don’t crash when they’re wrong. They just answer confidently and move on. That’s the failure mode traditional monitoring was never built for.

[Lucas Bolle | mibo-ai.com @lubolle]

OpenTelemetry’s GenAI conventions fix this: a common trace format for inputs, outputs, tool calls, and token usage across any framework.

[Lucas Bolle | mibo-ai.com @lubolle]

@saadscaleit We're building the reliability layer for AI agents — test them before you ship, then keep evaluating every real user trace after they're live. One test suite, two modes, zero guessing. https://t.co/TxMxZ2Sj2K

[Lucas Bolle | mibo-ai.com @lubolle]

@kushmergedeck Most teams ship AI agents and hope. We're building Mibo so you don't have to — same reliability tests running pre-release and passively on live traffic, without sitting in the request path. https://t.co/j7DU66fCn5

[Lucas Bolle | mibo-ai.com @lubolle]

Send a test request to a real agent and watch Mibo evaluate it live.

No login. No credit card. Pick a scenario, the agent responds, Mibo runs the tests and shows you the Reliability Score instantly.

https://t.co/j7DU66fCn5

[Lucas Bolle | mibo-ai.com @lubolle]

@LearnWithBishal You are right and most of the team just change the model, make some test and release it to the public.
To control if the agent is fail at a procedural level or at a semantic level we a friend we crate https://t.co/IpYG1eSK8H control tool calls, response type and tone and more.

[Lucas Bolle | mibo-ai.com @lubolle]

@ujjwalscript I completely agree with you, the hardest part is when the agent fail but looks like that everything is fine. For these reasons with a friend we create a SaaS to test this agents based on agent’s telemetry. If you want take a look to https://t.co/IpYG1eSK8H (100% free)

[Lucas Bolle | mibo-ai.com @lubolle]

@layla1136292 Exactly that is almost verbatim what happened to us with our original WhatsApp agent. We only found out when a user asked where their file went.

That gap between “looks fine” and “actually correct” is what Mibo is built to catch. Check it out https://t.co/IpYG1eSK8H

[Lucas Bolle | mibo-ai.com @lubolle]

We built a WhatsApp AI agent. You'd text it, send photos, it'd save them to your Drive, log data into Sheets.
We killed it 3 months later.
Not because of lack of users. Because of something worse 🧵

[Lucas Bolle | mibo-ai.com @lubolle]

@TheAIMargin Exactly. And you don’t find out from a dashboard for sure you find out from a churned customer or angry DM.
That is exactly why we built Mibo: catching “confidently wrong,” not just crashes. Curious if you’ve hit this with your own agents?

[Lucas Bolle | mibo-ai.com @lubolle]

If you're building agents on n8n, Flowise, or your own API, and you've got that feeling of "it works but I don't know why it sometimes doesn't" — this is exactly the problem we built Mibo to solve.
https://t.co/RUvAQAvIN2

[Lucas Bolle | mibo-ai.com @lubolle]

Mibo ships with a demo project built in, so you can see the Failure Matrix in action before connecting your own agent.

You can try it with no signup here 👇

[Lucas Bolle | mibo-ai.com @lubolle]

Amigos de @Newsan, si quieren que les demos una mano para que no pase más, ¡peguen el grito! 😉🚀
#BuildInPublic #AI #Testing

[Lucas Bolle | mibo-ai.com @lubolle]

La IA en producción sin testing es un peligro. Miren este loop de @Newsan: me escriben ellos, les digo que NO a una oferta y el bot flashea que quiero un humano para decirme que no hay nadie disponible. 💀
En mibo estamos para que estos flujos no mueran en el WhatsApp del cliente