About two years ago, I built an AI agent to have my own daily custom podcasts about any topic, like AI news or financial industry regulations.
Since then, I’ve kept improving it, testing new models and exploring what each advance in AI made possible.
What continues to amaze me is how much we can now connect and automate: researching stories, writing scripts, bringing AI presenters to life, producing audio and publishing the result.
I’m now sharing that experiment through I Wake Up, There Is New AI, a podcast and LinkedIn newsletter.
The objective is simple
A daily AI news podcast on YouTube. Multiple speakers reporting real, fresh news. A LinkedIn newsletter with the key stories, sources and a direct link to the episode. Subscribers can receive notifications when a new edition is published, according to their LinkedIn settings.
Everything, automated.
Building this workflow also gave me the opportunity to develop new architectures:
- Extending my agentic platform, Podcaas, with a ChatGPT plugin and an MCP server.
- Deploying GPUs on Google Cloud to run an open source Qwen AI model for voice cloning.
- Building a layered orchestration architecture where ChatGPT coordinates its own agents, operates the browser and also directs Claude’s agents.
A mini production studio, operated through conversation
The results were surprising. I no longer need my platform’s UI for this workflow. I can operate it by talking to it in ChatGPT, much like messaging a WhatsApp contact.
It is a real mini production studio that is almost entirely programmable. Almost still matters. This remains an experimental project, and I’m still refining the voices, editorial quality and reliability of the workflow.
A second verification changes the result, and the cost
One thing I learned: even when an article’s sources are included, a second verification can reveal differences in the facts when compared with other sources.
I added a second level of verification, using a verifier agent. This AI architecture pattern improved the quality of the output. But it also increased processing time, and therefore cloud cost.
That extra agent has a consequence. A better result also means more work for the system, more time and more resources.
Human engineering becomes even more valuable
There were plenty of mistakes along the way. Getting these components to work together reliably takes engineering.
For me, human engineering becomes even more valuable as we compose larger, more complex systems. This requires product and service architecture, and an understanding of each component we assemble. Each one has consequences for the value, quality, security and cost of the whole system.
This experiment has reinforced my conviction that software and infrastructure engineering have a bright future ahead.
It’s a way to follow AI’s progress while experiencing what it can already help us build.
Discover the podcast and subscribe to the newsletter:
Read and subscribe on LinkedIn · Watch on YouTube · Listen on Spotify
Listen on Podcaas · Subscribe via RSS
I just happen to like the meme the title originates from. And yes, you guessed right: this cat has the Steve Jobs pose.
Build, & learn
