I disappeared for months. This is what I was building

No, DevPleno isn't over. I went quiet for a while, but I didn't stop: over these months I was building products, testing lines of business and turning all of it into my own way of operating: starting and running several businesses practically on my own, with AI, automation and partnerships with specialists. This is the first essay from behind the scenes of that story.

The bottleneck was always me

Even with a team, everything went through me, and that capped how much could get done. Every decision, every review, every new project ran into the same queue. With AI the bottleneck doesn't go away, but it shrinks a lot. That's exactly why I can now run several fronts at once without any of them stalling while it waits for my calendar.

Alone in operations, partnered in the niche

The core idea: today you can do almost everything with AI (marketing, programming, operations), whether or not you're a programmer. But that creates a paradox. If anyone can "be an expert" in anything, what sets one business apart from another?

Hence the strategy: combine my experience in programming and orchestration with the expertise of someone who really knows the market. It isn't me asking AI about a market I don't know; it's the specialist setting the direction and using AI alongside. In practice, that turned into partnerships with management people, with a lawyer, and so on: several revenue lines, each with a niche partner.

And there's a great side effect: what I learn in one business feeds all the others. The billing reminder sequence born in one product becomes the standard in every product. The funnel that worked in one niche becomes a hypothesis in the rest. It's a compounding improvement loop.

Not everything needs to be an agent

A good share of the company's problems are solved today with agents and workflows, but with an important caveat. A lot of it is just a workflow with one decision point running an LLM. And a lot of it is still plain code: a billing reminder sequence doesn't need an agent, however nice it would look built with one. The flow that has been working: prototype the workflow in n8n and, once it proves itself, move it into code.

SaaS isn't dead

Plenty of people say SaaS is finished because anyone can build one now. Yes, anyone can build one, but people have other things to do. Could an accounting firm build its own CRM? Sure. But the accountant has her own operations to run. Two opportunities I've been exploring: infrastructure SaaS, which other businesses use as a foundation, and regulatory SaaS: new laws keep showing up that will require software. It's worth looking at what's appearing on the regulatory side of your field.

That reasoning is where this phase's products came from. PipesBot connects any application to WhatsApp. It was built first for an internal problem and became a product that former students already embed in their own SaaS. And Selva.run went after the two things that held up every new project: publishing content and sending email. Selva CMS is open source, multi-tenant and manageable over MCP: you can publish content through Claude. Selva Mail runs in your own AWS account, with costs that scale down to zero. The real gain is standardization: a new project means creating a tenant, pointing the domain, and it's live in a minute.

How do you learn to program now?

This is the part that unsettles me most. The field has changed a lot in the past year, and odds are we won't be writing nearly as much code anymore. If the act of programming changes, the way you learn it changes too. The best programmers today know how to specify: describe well what they want, not only the feature but how it should be implemented. And specifying takes concepts. That's why vibe coding gets so much criticism: people lack the vocabulary to ask properly and to judge whether the result is acceptable.

That's the challenge behind the new version of FullStack Master: rethinking the course for the AI era while keeping the concepts and the hands-on projects, but putting AI in the right place in the process instead of a copilot bolted onto the old course. There's no date yet, but that's the direction. And that's what this space will document: the whole process, not just the outcome.