Criador do bridgecode

Marcelo Santiago

Developer and creator of bridgecode, an agentic development environment for orchestrating squads of AI agents. Helps people going from vibe coding to professional development move from idea to shipped product.

GitHub Twitter / X LinkedIn Instagram Threads Website

Artigos publicados

252 artigos por Marcelo Santiago

Official OpenAI cover with an icon of a Git branch connected to three servers on a blue background, illustrating the article Scaling Git at OpenAI.

Newsletter

Nov 5, 20266 min

OpenAI builds Pando to scale Git with Codex

Pando is a Git compatible backend that OpenAI says it built to support the development volume of Codex in production.

#OpenAI#Pando#Git

Marcelo Santiago Read

Official OpenAI cover with the title A thousand fewer paths to root in white on an abstract blue and green background, referring to the research on the Linux kernel.

Newsletter

Nov 5, 20266 min

OpenAI reports 948 flaws in the Linux kernel

OpenAI describes a pipeline that validated findings in the Linux kernel and generated candidate patches for 948 privilege escalation flaws.

#OpenAI#Linux Kernel#segurança de IA

Marcelo Santiago Read

Example from the official README of the Villa Castagneto interface made with Impeccable, with a recorded nighttime scene and cream and red title.

Dicas

Nov 5, 20266 min

5 tips for reviewing interfaces with Impeccable

Use Impeccable commands to guide agents, review UX and detect visual patterns, but check permissions before enabling live mode.

#Impeccable#agentes de programação#design de interface

Marcelo Santiago Read

Official cover of the Hugging Face article about Phonon TTS, a speech model that runs locally.

Benchmarks

Nov 5, 20266 min

Phonon TTS tops local speech latency test

Hugging Face published on November 5 results for Phonon on the Open TTS Leaderboard, with 28 ms to first audio in the CPU streaming test.

#Phonon TTS#text-to-speech#Open TTS Leaderboard

Marcelo Santiago Read

Context Mode chart shows the median bytes of output that do not reach the model across five categories. The visible measurement is from November 1, 2026, version 2.0.0-next.23.

Dicas

Nov 5, 20267 min

5 tips to reduce context with Context Mode

Five tips for using Context Mode, filtering long agent outputs and measuring context savings, with license limits and security.

#Context Mode#agentes de programação#MCP

Marcelo Santiago Read

Official Images 2.6 artwork with a central blue card and generated image examples around it, including illustration, photo and informational layout.

Newsletter

Nov 5, 20266 min

Images 2.6 lands in ChatGPT with GIFs and editing

OpenAI launched Images 2.6 on November 4: faster generation, GIFs, more precise editing and new controls in the API.

#OpenAI#Images 2.6#GPT Image 2.6

Marcelo Santiago Read

Official BMAD Method banner on the GitHub repository

Dicas

Nov 5, 20265 min

5 tips for planning tasks with the BMAD Method

Use the BMAD Method to start with clear vision, short steps, acceptance criteria and a decision log.

#BMAD Method#dicas#planejamento

Marcelo Santiago Read

Official image of the CWE-bench v2 benchmark

Benchmarks

Nov 5, 20265 min

CWE-bench v2: method, ranking and limits

CWE-bench v2 opens 100 public tasks and reveals a leaderboard of 250 cases. See the method, the numbers and what the scoreboard does not measure.

#CWE-bench v2#CWE#benchmark de segurança

Marcelo Santiago Read

Banner of OpenAI's official showcase for community hardware, with computing devices and the phrase You can just build things with hardware.

Newsletter

Nov 5, 20266 min

OpenAI brings models into physical engineering

OpenAI describes models for CAD, simulation and hardware and an ongoing collaboration with Mazda. See the published results and their limits.

#OpenAI#engenharia física#CAD

Marcelo Santiago Read

Official Hugging Face illustration with a yellow character holding a code card and a validation mark.

Dicas

Nov 4, 20265 min

5 tips for testing models in Transformers with MIC

MIC organizes fixtures, usage paths and reviewed baselines to detect regressions in Transformers, without claiming tests that were not run.

#Transformers#Model Integration Contracts#MIC

Marcelo Santiago Read

Illustration of StateBench with colored lines that change position along a sequence.

Benchmarks

Nov 4, 20265 min

StateBench measures state tracking in long context

StateBench measures four levels of state tracking and shows how accuracy drops in long sequences, with explicit methods and limitations.

#StateBench#Ai2#benchmark

Marcelo Santiago Read

Diagram of the Dharma Field Extractor showing a purchase receipt on the left and structured fields in JSON on the right.

Newsletter

Nov 4, 20265 min

Dharma Field Extractor turns documents into JSON

Dharma Field Extractor turns documents into JSON fields. See the training, the published evaluation and what still needs validation.

#Dharma Field Extractor#Dharma-AI#OCR

Marcelo Santiago Read

The biggest AI newsletter in the World

Follow the evolution of AI squads and multi-agent systems.

In-depth technical curation: agentic architectures, real token benchmarks, MCP servers and open-source tools in your inbox and on the blog.